Noxa indicators for Neuroshell

Currently I am evaluating the EigenVectors indicator in Noxa. It is used to define the GroupStart and GroupDepth for some Noxa indicators parameters used in Trading Strategy.

Some Lissajous pattern formed differently when shifting the Eigenvectors. Anybody can advice which Lissajous pattern should be selected to provide good trading strategy?

Here samples of Lissajous pattern formed at different GroupStart and GroupDepth values and the respective trading result, the chart with data is inserted on zip file.

On the help it mentioned "The corresponding Lissajous pattern is a circle. Their inherent persistence over time makes them especially worthy of attention." But to get the circle pattern is not easy.

Cheers,
Arry

Hello Arry,

I had trouble using the Eigenvectors too. In fact they dumped it from the Beta version that I am testing for MetaStock.
Anyway I found the eigenvectors more useful on indicators rather than on price; any indicator for that matter. I look more precisely for a plateau in the spectrum (the wider and the flatter the best) where several pairs of eigenvectors plot a circular Lissajous. The eigenvectors in a pair appear shifted indicating a prediction mode…

Ps. Your optimization came out with about only 100 bars for the training range and a m-histories of 50. I don’t know much about Forex but your eigenvectors look pretty noisy; I wonder whether you can get anything good from this model. I would have expected something like m-histories = 250 or more with a training range at least four times that!

My two cents.

P.Z

Have anyone tried to predict cycles?

I could write a thesis predicting cycles!

I got very mitigated results trying to predict CSSA and even cycles using NeuroShell. I think that FeedForward neural nets cannot deal well with temporal sequences. I did not give up and developed my own models with some encouraging results but I was barely able to predict more than one full period ahead. However I got some very good results when trying to predict SSA with CSSA as inputs with a memory-based recurrent state space model. Now my predictions go beyond 6 periods with good accuracy of the turning points! There is something very interesting to learn from the CSSA/SSA relationship!

P.Z

Nice idea, will have to try it out. I had no success predicting SSA before. However, I tried to predict MA Momentum feeding Fourier harmonics as inputs. Basically I was trying to find harmonics presented for some period of time in the market. It appeared to be good at training data and not so promising on out of sample data. Did you try to feed correlation along with inputs? Also, what software did you use? Or is it self-made?

Hi Sedkol,

I have attached my first good results of SSA prediction for you to see. The prediction starts at bar# 2450. The objective is to predict the next 50 bars as they are missing in the original SSA (m-histories=50).

Yes the software is mine. Just to give you a hint, I first use a Genetic Programming and some artificial embryogeny to grow a state space machine from the in-sample data. Then I train my machine using several rules inspired from biology. Finally I let the machine generate the cycle by itself. This is that simple '-)

As for using correlation? I tried but unfortunately my results were not very robust. I extensively use the cointegration metric form Granger though: http://repositories.cdlib.org/cgi/vi...ntext=ucsdecon

Remember correlation is not causation!

P.Z

From what I can see, P.Z, you had quiet stunning results predicting SSA in terms of causation. Keep up the great posts, I am more and more into that thing and really think your posts are very helpful.

Sedkol,

I found trading excessively competitive; there is no such thing as a software that will make you happy by casually clicking on green and red arrows with a smile on your face. I bet the lifetime of traders is less than a few months for most of us because the (scam?) trading industry sell us dreams. This is hard work to develop strong principles that work. But I found the guys at Ward Systems and Noxa very inspiring and now that it started paying off I thought I should give back some of what I learned the last 15 years...

P.Z