Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Neural network trading 34 replies

Neural Network Trading with Simbrain? 0 replies

Neural Network trading using Matlab and Metatrader 8 replies

Interested in Neural Network Assisted Trading 4 replies

neural network based systems in forex 22 replies

Attachments: Neural Network Trading, Serious people only!
Exit Attachments

Neural Network Trading, Serious people only!

Quoting fajst_k
So for the same period (1 day out of sample) standard net earned 39 pips and recurrent 112 pips (both comission 3 pips). Paper trading results were also much better for recurrent net. Input data was low quality, so possible space for improvement is big here by feeding the net with high quality training and optimalization data to obtain stronger non linearity.

Krzysztof
How well does it work on more than 3 trades?
I got a template in NShell, it look very profitable but I'm having problems sending odrer to Mt4, Here is the system screen shoot at FreeImageHosting.net Hosting Service and FreeImageHosting.net Hosting Service. I'm using Tradeput EA in MT4 here is the shot at FreeImageHosting.net Hosting Service

Attach is the template

Thanks
Attached File(s)
File Type: zip TS sasà 15 min + Tstop40 lyvyo.zip   362 KB | 1,781 downloads
Quoting fajst_k
Hi,

Here is an advanced strategy for NS, it uses JMAs crossovers, HP filter and Ehlers trend line. It uses also LEAD function, the usage of this function in real time trading is unclear for me.

Included chart has 3 Strategies, JMACROSOVER, FIlter1 and FIlter2, it you have a problem with accessing existing data calculations or adding an alarm, it is necessary to extract all strategies separatelly and put to the new chart.

Krzysztof

Can somebody help me with this JMACROSSOVER.ZIP
i can download but i can't extrack this files
Thank's ...
Quoting Andelias
Can somebody help me with this JMACROSSOVER.ZIP
i can download but i can't extrack this files
Thank's ...

Use latest winzip

Krzysztof
Quoting fajst_k
Use latest winzip

Krzysztof

Thank's a lot Krzysztof, for very quick respon.

http://www.forexfactory.com/images/icons/icon14.gifhttp://www.forexfactory.com/images/icons/icon10.gif

Andelias
Anyone had a look at this opensource neuron network library?
http://www.cimne.com/flood/
Is it possible to reduce the indicators dimensionality with neural network(kohonen etc..)without facing the problem of overfitting or multi colinearity? ? I mean for example, lets throw RSI , CCI and pivot/fibo indicators in the chart, when RSI , CCI and pivot/fibo indicators agree is it possible find optimal overbought-oversold area with NN ?
Hi,
I am interested in this area, but have no idea how to begin. Can anybody recommend an EA that does this evaluation?
Ron
Hi everyone,

First of all, congratulations for the work so far! But I would suggest we try these archtectures on Matlab (or R) first before going through all the pain of using the FANN in MT4. Anyone with me?

I have a nice statistical background but I`m a newbie in ANN. I`m trying to reproduce on Matlab the results found here

I have a set of inputs, 1000 daily data points, the last being taken at midnight of yesterday, suppose it is 00:01 am and i want to predict the returns (based on open and close) for today.

my input consists of open, close, min, max, volume, log returns (for 1,2,3,5, and 8 periods) based on open and close, ROC 14, and ADX 14.

my target is a matrix with 2 rows, one of log returns based on close and the other based on open.

I did NOT lag the input by my hand. I'm using the following matlab command:
net=newnarx(input,target,[0,1],5);

My question is:
Does this network predict the series one step early? I mean, if I run this at 00:01 am i'll be estimating the log returns for today?

I'm really looking forward to working with ppl from here.

Btw, did anyone read about high order neural network? Looks quite promising also.
Quoting lgbpinho
Hi everyone,

First of all, congratulations for the work so far! But I would suggest we try these archtectures on Matlab (or R) first before going through all the pain of using the FANN in MT4. Anyone with me?

I have a nice statistical background but I`m a newbie in ANN. I`m trying to reproduce on Matlab the results found here

I have a set of inputs, 1000 daily data points, the last being taken at...

I've been using FANN for awhile. It's not to bad. Matlab is expensive unless your a student.

Not sure what your doing here. Usually with an ANN you start with a set of training data. For example,

day+1 Close, open, high, low, close, etc.

The first data point you would specify to the ANN as output(predictive) and the rest as input. After the ANN has been trained, you just use the input and it gives an output.

Many people, including me have tried to predict based upon what you say your using with no luck. To make an ANN work, you need to help it out a bit by representing the relationships between data. So for Open, high, low close it maybe better to use open-high, open-low, open-close. It doesn't know what the implied relationship between a high and low is, to the ANN it's just data. Success is all in the data.

If I was to start again, I'd primarily look at the problem from the perspective of data mining. Find the data that meaningfully stands on it's own, then create an ANN.

Good luck!
.
first post here..

hi guys ..everyone

thanks for who created this thread..charlinks

yes for serious guys who make money in a million

I also love NN style..

nature forcing of waves in my simple words

using the main Algorithm formula based on human brain's cells

Xi = Xi - Xmin
________

Xmax - Xmin

with viral methodology for outlook
2 types of prediction error ESM NN

was so great..

Im so serious studied this subject




ps must love formula, math , adv math.. logical thinking...


this is one of the formula for exponential of NN and forecasting prediction R X chart of NN
Attached Images
Quoting redpine
I've been using FANN for awhile....
Well, yeah, FANN is not bad, actually it is good, I think that my experience with it was bad because i'm not much of a programmer.

Besides Matlab (which is really expensive, i use the one at my university), we can use R, which is free . My point is, it is much easier to test a NN and trading strategies in softwares like R and you can know beforehand if it is likely to work or not, before implementing.

So, yeah, the basic start point is to investigate the data, I've got a bunch of technical indicators and some 1500 data points for a principal components analysis (requires lots of data) so i can narrow the dimensionality of the inputs to less than 20. Then i trained this network with a recurrent hidden layer, as described in the paper i post last thread. Now i need to calibrate it for another pair, prolly using some genetic algorithm, then i measure the quality of the network and theeeen i try to use it for real.

Do you know any nice FANN tutorial? Maybe i would still use MAtlab or R to make the predictions instead of having an EA... i use daily data only (high frequency is tough :S)
Quoting lgbpinho
Well, yeah, FANN is not bad, actually it is good, I think that my experience with it was bad because i'm not much of a programmer.

Besides Matlab (which is really expensive, i use the one at my university), we can use R, which is free . My point is, it is much easier to test a NN and trading strategies in softwares like R and you can know beforehand if it is likely to work or not, before implementing.

So, yeah, the basic start point is to investigate the data, I've got a bunch of technical indicators and some 1500 data points for a principal...
I mainly use FANN because I'm C++/Java developer in my day job. With FANN I've been able to implement some the Reinforcement learning algorithms the same author created and use them. I also use a product called Beagle which gives some good implementations of Genetic algorithms and Genetic programing.

You can run similar products with EAs or Java or VB, but my datasets tend to be very large ~350GB. I've currently have one year of L1, L2 and Time Series data stored in a MySQL database. So again I need something that can perform fast and efficiently. This is why I choose C++ and FANN.

It I had MatLab, I'd use it instead. R I never really spent the time to learn.

FANN is pretty easy to use. There is a GUI for training and setting up the data is easy. It also has a feature I use called Cascade Training, which figures out the optimal layers and nodes.

The biggest problem with ANNs in trading is that most solutions are just curve fitting. The Genetic algorithms also suffer from this. The key to a predictive Neural Net will be in how you prep the data. TA representations of data don't work. I've tried most everything and always end up with a curve fitted solution.

My current spin is Q learning. I see the market as a finite state machine and would like to use Q learning to predict the next state. But I'm still thinking this through. :-(
.
I've been studying a lot lately, and finally got some clue of how this thing works (finally). My question now is, using which inputs did you guys get success? From the books and papers i read, the prices or (exclusive or) the returns are enough. I was able to use log-returns obtained from opening to opening to predict log-returns based from minimum to minimum with a very nice fit, which is still worthless but at least it is something.

So, which inputs and architecture did you have the most success with?
The Spinal Implant EA is Free and open source unsupervised Neural Net Scalper with Fractal Geometry filtering. This is an implementation of unsupervised neural net with Genetic algorithm optimization by MT4.

It is based on the Artificial intelligence by Reshetov but I use non linear activation function and fractal dimension filtering.

in the original:
double w1 = x1 - 100; by double w1 = x1/1+MathAbs(x1);

The idea is to use the native genetic algorithm of MT4 in order to look for a solution this time with non linear activation function: x/(1+|x|).

The fractal geometry filtering is used in order to filter out the shitty range conditions.

the original code
double w1 = x1 - 100;
double w2 = x2 - 100;
double w3 = x3 - 100;
double w4 = x4 - 100;
double a1 = iAC(Symbol(), 0, 0);
double a2 = iAC(Symbol(), 0, 7);
double a3 = iAC(Symbol(), 0, 14);
double a4 = iAC(Symbol(), 0, 21);
return(w1 * a1 + w2 * a2 + w3 * a3 + w4 * a4);


The code of the perceptron with rational activation function x/(1+|x|).

It is a good practice to use less neurons when you use a genetic optimization. (Otherwise you will need to optimize forever LOL).

OK here it is:

double perceptron()
{
double w1 = x1/1+MathAbs(x1);
double w2 = x2/1+MathAbs(x2);
double w3 = x3/1+MathAbs(x3);
double w4 = x4/1+MathAbs(x4);

double p1 = iCustom(NULL,Symbol(),"PFE",x5,true,5);
double p2 = iCustom(NULL,Symbol(),"PFE",x6,true,5);
double p3 = iCustom(NULL,Symbol(),"PFE",x7,true,5);
double p4 = iCustom(NULL,Symbol(),"PFE",x8,true,5);

return(w1 * p1 + w2 * p2 + w3 * p3 + w4*p4);
}

For copy right reasons the EA is not shared here. This is not commercial software and definitely is not plug and play.

If you like it, there is a like button LOL.
Attached Image (click to enlarge)
Click to Enlarge Name: fractal slalper.png Size: 276 KB
Attached Image
Hi all.
I am new to this forum but I had many trade related posts in a persian forum.

I used neural networks for trading for years and I designed and implemented them by myself. I used support vector machines too.

My results are not satisfying yet. There is times that I think that machine can do nothing with Forex trading but something in my soul makes me hopeful.

I spent thousands of hours working on automated Forex prediction. I should mention that I have PHD. in computer science and I am doing serious programming for past 10 years.

After all, my question is that:

Can a machine be a successful trader in long time. (I believe that for short time there can be success - which I thing an accident).
Quoting arryex
Hi Friends,

I give you here attached free e-book (one of my collection) which I created based on my own experience using Neuroshell. This e-book may useful for Neuroshell users.

Any comments are appreciated.

Have a nice reading,
Arry
Can you please post how to connect MT4's DDE feed into Neuroshell? That would be helpful.
  • Post #118
  • Quote
  • Edited 12:15pm Jun 30, 2012 12:12pm | Edited 12:15pm
Quoting mwduke
Can you please post how to connect MT4's DDE feed into Neuroshell? That would be helpful.

There was a guy who used to do it, but he was a flake and didn't support his product. Turns out it was faulty.
I went back and forth with him and Ward Systems.
They (Ward systems) were great, but in the end they ended up severing ties with the guy.
You can go to neurshell.com and look under compatible data feeds. That's where it used to be.

Best bet is to contact them for assistance. I don't use MT4 for a trading platform, only for charting and I'm going to be ditching MT4 all together very soon.
Here's the latest article from Futures Magazine: Casting neural nets into modern markets.
Attached File(s)
File Type: pdf Casting neural nets into modern markets.pdf   282 KB | 1,515 downloads
Does anyone know how to add custom indicators from MT4 (in mq4 format) into Neuroshell? Like Ichimoku?


or also how to create custom periodicities/time frames?
0 traders viewing now