Build Neural Network Indicator in MT4 using Neuroshell

[Krzysiaczek99]

Interesting comment from Peltarion Team '' SOMs and SVMs only work on static algorithms'' so what types of NN and SVM algos work for continous time series?

Well theories are different about it. Thanks to our Chinese brother we have this
collection of papers which are researching this a bit

/trunk/papers/trading/ - zenithlib - library everywhere - Google Project Hosting

svn checkout zenithlib - Revision 119: /trunk zenithlib-read-only

Krzysztof

 

[Krzysiaczek99]

Well theories are different about it. Thanks to our Chinese brother we have this
collection of papers which are researching this a bit

/trunk/papers/trading/ - zenithlib - library everywhere - Google Project Hosting

svn checkout zenithlib - Revision 119: /trunk zenithlib-read-only

Krzysztof

Wow !!! the paper from the library form link above

2003-Using_Recurrent_Neural_Networks_To_Forecasting_of_Forex

says that.


Probability of correct sign of increments forecasting for CHF EUR GBP JPY
0,79 0,78 0,78 0,89

Otherwise very good straight forward paper.

Krzysztof

 

[Krzysiaczek99]

Hi All,

This article shows how Kohonen SOM implementation using Viscovery SOMine.

My conclusion is this SOM classification help me out to select some good stocks and then use technical analysis as my trade entry/exit.

Good tool for this is VisuMap also

VisuMap Technologies: Visualizing high dimensional complex data

 

[arryex]

Hi Kryzstof,

Long time not see the thread...a lot of nice things you put here.. Thanks..

After registration, link to download Visumap is http://www.visumap.net/registered/VisuMap3_5.msi
inside via this link http://www.visumap.net/registered/RegMain.aspx

Documentation found here http://www.visumap.net/registered/Document/VisuMap/VisuMap.html

 

[arryex]

Recent updates...
- Transfer MT4 indicator to NST
- Improve indicator in NST

 

[arryex]

In this article I use K-Means Clustering in Visumap to classify stocks. The Visumap result is inline with my previous clustering using Kohonen Map in Neuroshell2.

I used clustering as a preliminary analysis prior selecting stocks that will be analyzed using trading strategy in Neuroshell Trader.

Anyone can check is it worth to have these software or not.

Happy Investing
===============================
I am an engineer not scientist, straight forward real implementation is better than thousands theories

 

[tovim]

Arryex

I have not fully investigated Visumap,addition to classic clustering algorithms data mapping(dimension reduction)algorithms looks like great addition .Do you or Krzysiaczek know how to fully take an advantage of data mapping ?

 

[arryex]

Hi Tovim,

Refer to help of Visumap, it has capability to export/import data, mapping data, clustering data and interactive data view.

For data mapping, it has several algorithms: Sammon, CCA, RPM, Speric Relational, PCA, SMACOF and t-SNE. For clustering, it use K-Mean, SOM, metric sampling, Agglomerative, Affinity and SOG.

A lot of data mapping and clustering algorithms available, sometime not available in one software and available in others, the important thinks is "Our capability to evaluate/interpret produced cluster or map". The clustering and map will be useless if you do not knows how to interpret them. The method I took to evaluate them is by testing each cluster into my trading strategy.

Within attached article (ultsch09selorgan.pdf), shown other algorithm that is not available in Visumap, example SOP (Swarm-Organized Projection).

Within attached article (NA16409.pdf), it shown the SOM visualization comparison using Nenet, Matlab SOM-Toolbox, Databionic ESOM and Viscovery SOMine. In chapter 4, the author explains few difficulties to interpret the map which lead him to create a new method and then implement it on Matlab.

I found that Sammon Algorithm publicity has been released since May 1969, Kohonen SOM in 1982, CCA in 1997, RPM in 2004. I agree to mention that Visumap has more algorithm while other may have few algorithm only, and again it will depend on our capability to evaluate/interpret produced cluster or map.

Happy Investing

 

[seriousjoker]

hi folks . i m trying to use SVM but i ve encountered some problem in deciding the output for the training set. How do you usually define a buy/sell/signal ? like the next 3 days return exceed 0.5% ?
i m using technical indicator values as input
thanks very much

 

[Krzysiaczek99]

In this article I use K-Means Clustering in Visumap to classify stocks. The Visumap result is inline with my previous clustering using Kohonen Map in Neuroshell2.

I used clustering as a preliminary analysis prior selecting stocks that will be analyzed using trading strategy in Neuroshell Trader.

Anyone can check is it worth to have these software or not.

Happy Investing
===============================
I am an engineer not scientist, straight forward real implementation is better than thousands theories

Clustering of time series is a big topic. Perhaps this pdf can show a bit what's going on in this area.

Krzysztof

 

[arryex]

hi folks . i m trying to use SVM but i ve encountered some problem in deciding the output for the training set. How do you usually define a buy/sell/signal ? like the next 3 days return exceed 0.5% ?
i m using technical indicator values as input
thanks very much

If you have Neuroshell Trader, you can insert indicator from Neural Output group to get optimal Buy/Sell/Hold.

 

[arryex]

Hi All,

Anyone using Wolfe Wave to predict target price and time? I did on few of stocks

 

[Krzysiaczek99]

Multicollinearity and forecasting

Are we doing all this stuff correct ?? We create predicting variables based on one input i.e. historical time series, most of them are very strongly correlated to each others. Perhaps other inputs e.g. from different time series should be used instead of pre processing one serie using e.g indicators. See 'Multicollinearity and forecasting' chapter.

5/7 Correlation, causation and forecasting | Forecasting: principles and practice

On the other hand i tried this some time ago (4 series) and it didn't improve the results. Did somebody tried to use non correlated inputs and cross check the results ??

The online book 'Forecasting: principles and practice' is very good, soon i will open the new thread to research all those methods. If someone is aware about similiar threads (e.g. ARIMA, GARCH, ETS) and attempt of using them for FOREX let me know.

Krzysztof

 

[arryex]

Probably this small article may help us to identify which correlation coefficient we should use on prediction

 

[Krzysiaczek99]

is FOREX forecastable ?

is FOREX forecastable ?

according to main guru Rob J Hyndman in forecasting of time series not so much...

Research tips - Why are some things easier to forecast than others?

My opinion is that it is forecastable but only sometimes on short time horizon. My opinion i'm basing on fact that it is possible to optimize strategy parameters that it will work for a while sometimes....

Any comments or articles ??

Krzysztof

 

[tovim]

This paper is ambitious about forecasting exchange rates :

'' Hidden Markov foreign exchange rate forecasting model (HFERFM) showed an accuracy of 69.9% as against 81.2% accuracy of AFERFM. This shows that the new approach provided an improved technique for carrying out foreign exchange rate forecasting.''

http://www.wcsit.org/pub/2011/april...del for Forecasting Foreign Exchange Rate.pdf

 

[arryex]

I am currently evaluating NN prediction using one of NST addon implemented on few stocks.

 

[arryex]

Here is the testing result on stock, EURUSD and XAUUSD

 

[arryex]

I did the following charts within an hour, not sure if I am ambitious

 

[arryex]

I just playing with GMDH polynomial net, few useful links:
Forecasting Software for Professionals (GMDH shell, I use trial version)
Group Method of Data Handling (GMDH) for data mining, forecasting algorithms optimization, fuzzy models analysis, statistical learning networks and modeling software systems

I made a simple time series prediction using GMDH with GMDHShell, NS2 and Matlab, see the detail within article.