I would like to answer your question but main idea was to share the fractal break - out ideas.
I have used Neuroshell Day Trader with the metatrader datafeed. What I like with it. I like that I can implement really quickly trading ideas.
What I do not like. I do not like that it is a black box, you have to pay for, well some people do not but this is another story . Of course I know that it uses some kind of proprietary cascade correlation algorithm. They call that turboprop, and it is maybe a cousin of the quick-prop algorithm.
And the neural net have limitations (of course).
If you are interested in investing in it, I can recommend the Products of Noxa analytics as the CSSA (this product was commented a lot here and other treads).
I want to emphasize here its relation with the fractal dimension. A member has observed that the CSSA give a lot of good signals in particularly noisy environments but when a sharp shift in the market occurs the CSSA loose the fit.
So there come the idea to constantly monitor the market with the fractal index and I know that such a shift occurs when there are a strong and violent variation of the Hurst exponent. So you use your net but you have a panic button.
My observation of the Neuroshell and the other nets is that they really cannot predict break - outs. So my opinion the best use of the neural nets are in relatively calm market environments when a break - out is unlikely to occur. You learn very well your retail provider algorithm LOL.
I think a lot of neural net practitioners have faced how a neural net counters the trend .
In fact you may laugh at my reductionist approach but a neural net in my opinion is just a better and nonlinear indicator. It looks like an advanced non linear oscillator (in fact the Neuroshell guys do not try to predict price levels, they predict percentage change in open).
I think the best that can be done on a retail level is to combine SSA with Support Vector Machines and use it in a calm market environment. There are several open source softwares that permit that combination. That will be my range bound market system.
If the price series are clearly persistent with medium to strong volatility I would prefer a digital filter optimized with genetic algorithm.
I have used Neuroshell Day Trader with the metatrader datafeed. What I like with it. I like that I can implement really quickly trading ideas.
What I do not like. I do not like that it is a black box, you have to pay for, well some people do not but this is another story . Of course I know that it uses some kind of proprietary cascade correlation algorithm. They call that turboprop, and it is maybe a cousin of the quick-prop algorithm.
And the neural net have limitations (of course).
If you are interested in investing in it, I can recommend the Products of Noxa analytics as the CSSA (this product was commented a lot here and other treads).
I want to emphasize here its relation with the fractal dimension. A member has observed that the CSSA give a lot of good signals in particularly noisy environments but when a sharp shift in the market occurs the CSSA loose the fit.
So there come the idea to constantly monitor the market with the fractal index and I know that such a shift occurs when there are a strong and violent variation of the Hurst exponent. So you use your net but you have a panic button.
My observation of the Neuroshell and the other nets is that they really cannot predict break - outs. So my opinion the best use of the neural nets are in relatively calm market environments when a break - out is unlikely to occur. You learn very well your retail provider algorithm LOL.
I think a lot of neural net practitioners have faced how a neural net counters the trend .
In fact you may laugh at my reductionist approach but a neural net in my opinion is just a better and nonlinear indicator. It looks like an advanced non linear oscillator (in fact the Neuroshell guys do not try to predict price levels, they predict percentage change in open).
I think the best that can be done on a retail level is to combine SSA with Support Vector Machines and use it in a calm market environment. There are several open source softwares that permit that combination. That will be my range bound market system.
If the price series are clearly persistent with medium to strong volatility I would prefer a digital filter optimized with genetic algorithm.