Build Neural Network Indicator in MT4 using Neuroshell

Dear All,

As previously stated that I will shown you how Chaos Hunter generating signal within NSDT, here I put my own trading signal. Currently my real trading profit is almost increasing more than 14%.

You may use as well MT4 data if you want.

regards,
Arryex
www.ai-traders.org
 

Attachments

  • Auto.JK using Chaos Hunter.png
    Auto.JK using Chaos Hunter.png
    24.8 KB · Views: 2,028
  • BBRI.JK using Chaos Hunter.png
    BBRI.JK using Chaos Hunter.png
    35.5 KB · Views: 1,077
  • BBRI.JK Chaos Hunter Optimization.png
    BBRI.JK Chaos Hunter Optimization.png
    60 KB · Views: 784
  • AUTO.JK Chaos Hunter Optimization.png
    AUTO.JK Chaos Hunter Optimization.png
    43.5 KB · Views: 957
  • AUTO & BBRI.png
    AUTO & BBRI.png
    9 KB · Views: 375
Hi

I bought VantagePoint for USD 2500 and am not able to make money it. After some time I gave up to trade the predicted highs and lows because it was just so wrong.
I really liked the idea but the problem is, that this software does not work in real time.

Thus, I got Neuroshell Trader and also a MT4 NN indicator which i am currently testing.
But if you are not much of a developer NS might not be the answer.
There are other NN tools for Tradestation and Ninjatrader which I might try.
In general, the tools for MT are completely inferior to the other two platforms. Just look at the prices they are able to charge. I think people waste their time in insisting in using MT. Unfortunately, it is still the most practical solution for FX....



Anybody know about VantagePoint's Intermarket Anaylisis?

As explained in the book "Trend Forecasting with Technical Analysis" by Louis Mendelsohn, intermarket analysis provide more precise trading strategies.

Based on given inputs, there 5 Neural Networks are produced: 1. predicted 1 bar high (PHigh), 2. predicted 1 bar low (Plow), 3. predicted short term 2 bar SMA5 (PTS), 4. predicted medium term 4 bars SMA10 (PTM), 5. Neural Index (based on difference between predicted 1bar and actual SMA3).

Additional indicators PTS Diff (PTS-actual SMA5) and PTM Diff (PTM-actual SMA10), PHigh Diff (PHigh - High) and PLow Diff(PLow-Low). Unfortunately, I could not found how to get "Strength Index".

Neural Network Inputs: for EURUSD chart daily.
Open,High,Low,Close,Volume,AUDJPY, AUDUSD, EURCAD,GBPJPY, GBPUSD,,^DJI, ^DJT, ^DJU, Nasdaq-100, FTSE 100. I could not include Oil, USD Index, T-Notes as mentioned in the book, due to unavailability data from MT4.

Neural Network outputs:
Lead(High,1), Lead(Low,1), Lead(Avg(Close,5),2), Lead(Avg(Close,10),4) Lead(Avg(Close,3),1)

NN design using back propagation.
The book explain how to use these indicators, such as:
- Buy when PTS Diff or PTM Diff turns positive and sell when PTS DIff or PTM Diff turns negative.
- Buy when PTS cross above TrendS or PTM cross above TrendM and sell when reverse
- Buy when PTS cross above PTM and sell when reverse
- Buy when Neural index 1 and sell when 0.

Higher probability to be used for trading, i.e. when all indicators are in confluence.
 
Hi

I bought VantagePoint for USD 2500 and am not able to make money it. After some time I gave up to trade the predicted highs and lows because it was just so wrong.
I really liked the idea but the problem is, that this software does not work in real time.

Thus, I got Neuroshell Trader and also a MT4 NN indicator which i am currently testing.
But if you are not much of a developer NS might not be the answer.
There are other NN tools for Tradestation and Ninjatrader which I might try.
In general, the tools for MT are completely inferior to the other two platforms. Just look at the prices they are able to charge. I think people waste their time in insisting in using MT. Unfortunately, it is still the most practical solution for FX....

Vantagepoint is a total waste of time, myself and several others worked on testing it for well over a year but it did not give an edge at all. I dont actually believe it does intermarket analysis because I was able to replicate the so called 'neural index' using simple calculations based on OHLC prices. I'm not saying that intermarket analysis itself is useless, merely that Vantagepoint is useless.
 
I have a problem using neuroshell
when the conditions set ,for opening a buy order for example,are met ,neuroshell executes the order at the open of the next candle
but I want it to execute the order immediately when conditions are met ..in the same current candle
anyone knows how?
 
I had made few correspondings with Neuroshell people in the past and they said that PNN with GA for classification GRNN with GA prediction is not exactly suitable for trading.On the other hand they claim them as best prediction and classification NN algorithms.Has anyone tried these two NN algos for trading purpose?
 
I've done quite a bit of investigation with PNN and GRNN with financial time series. I prefer simple single hidden layer networks with only a couple of hidden neurons and no more than four inputs. Nets are really good for building models for well behaved (ie low noise/consistent) time series but as soon as the data gets jumpy too many neurons produce too lumpy a surface.
 
hi all.. i just joined this forum cuz of this really good thread.. hopefully would contribute something to the community as well.. :)

ive checked out the first post in this thread and the attachments along with it.. ive got a couple of questions regarding it that i cant seem to get a hang of it..

1. the "data collector script" creates a *.pat file with 300 samples whereas the *.pat file already included in the package contains 5000 samples. why this discrepancy?

2. the *.pat files have the same identical samples. you can check this out in the mq4 code too.. the technical indicator functions used in the main for loop use hard-coded shift values.. for example..
  • input[3] = gmThreshold(iRSI(NULL,PERIOD_M15,9,PRICE_CLOSE,4)/100);
instead of something like
  • input[3] = gmThreshold(iRSI(NULL,PERIOD_M15,9,PRICE_CLOSE,4+cnt)/100);
where cnt is the loop counter.. this in essence means that we intend to train the network to just this sample.. why would we wanna do that?

3. ive modified the code to include the counter variable in indicator calculations.. i use the open-source fann library and its open-source gui fann-tool to train and test networks.. for 5000 data points, the mse doesnt seem to fall below 0.0045.. plotting the trained network's output against the training output shows the network hasnt generalized that well at all even at this mse.. the question.. what sort of mse should i be aiming for and which tools have you guys used to train the network?

thanks a million for your help in advance guys.. love this thread!!! :)

asim..
 
Vantagepoint is a total waste of time, myself and several others worked on testing it for well over a year but it did not give an edge at all. I dont actually believe it does intermarket analysis because I was able to replicate the so called 'neural index' using simple calculations based on OHLC prices. I'm not saying that intermarket analysis itself is useless, merely that Vantagepoint is useless.

Wow, I wouldn´t call it like this but certainly there are other systems and tools which may get you there more easily.
I cannot imagine that it is not doing inter-market analysis as I see the included asset classes in the prediction but of course this could be fake...everything is possible in this world....
You are right, inter-market analysis is definitely important and I believe it has never been more important. Larry Williams` Will indicator by the way is an simple example for inter-market analysis. I wonder whether this could be an interesting input in a Neural Net.
 
Hi Dergham,

Probably your input for your trading signal are using current opening not the previous close, hence the signal wil lbe generated if current candle is completed.

I think you can use prediction strategy and use predicted future value as your trading signal. Then the signal will be generated based on some previous candle and no need to wait your recent candle to be completed.

Arryex

I have a problem using neuroshell
when the conditions set ,for opening a buy order for example,are met ,neuroshell executes the order at the open of the next candle
but I want it to execute the order immediately when conditions are met ..in the same current candle
anyone knows how?
 
Hi Tovim,

I am following idea that previously created on Thesis made by Erin S Heim, "A neural network approach to the prediction and confidence assignation of non linear time series classification". You can download freely this ebook from the net.

The idea is to make a prediction of S&P100 daily change and some indicators as inputs. He made two neural networks:
- Using back propagation to find predicted change 12 inputs, 1 hidden layer and 1 output.
- using Kohonen network to classify inputs and generate confidence factor, 12 inputs indicator with 30 output classification. Based on how many percentage of correct classification (Buy/Sell) with certain input condition we define the confidence factor. I get some class for Buy with minimum 75% up to 100% correct, and remove all percentage below 70%.

Afterward we decide to go for Buy if the predicted daily change is positive and the confidence factor is above 75% to buy, and Sell in reverse conditions (negative change and confidence factor above 75% to sell).

I tried implementing his idea and looks giving better trading performance. Sometime I will post the NN description and the result.

What I would like to conclude is, we can utilize all the available network to support our trading decision. In this sample I use back propagation to predict and Kohonen to classify and use both of them to support your trading decision.

By explaining this idea, hope you have other idea to combine PNN to classify and GRRN to predict, then use to support trading decision..and I will...why not?

Arryex
I had made few correspondings with Neuroshell people in the past and they said that PNN with GA for classification GRNN with GA prediction is not exactly suitable for trading.On the other hand they claim them as best prediction and classification NN algorithms.Has anyone tried these two NN algos for trading purpose?
 
Hi Awesim,

I try to answer your question:
1. The data script can be made adjustable depend on the date we assign after, I use it as extern variable so you can change as you want.

2. With the code I use, the data will be written as the counter change, hence you get the time (the counter), the indicator values then write it into csv file. I do not think with counter inside you can write the file at the same time.

3. Personally I use neuroshell2 as my tools but nevertheless to use other tools such as neurosolution, chaos hunter, neuroshell predictor/classifier. I do not use other tools.

Sorry for the late repply.
Arryex

hi all.. i just joined this forum cuz of this really good thread.. hopefully would contribute something to the community as well.. :)

ive checked out the first post in this thread and the attachments along with it.. ive got a couple of questions regarding it that i cant seem to get a hang of it..

1. the "data collector script" creates a *.pat file with 300 samples whereas the *.pat file already included in the package contains 5000 samples. why this discrepancy?

2. the *.pat files have the same identical samples. you can check this out in the mq4 code too.. the technical indicator functions used in the main for loop use hard-coded shift values.. for example..
  • input[3] = gmThreshold(iRSI(NULL,PERIOD_M15,9,PRICE_CLOSE,4)/100);
instead of something like
  • input[3] = gmThreshold(iRSI(NULL,PERIOD_M15,9,PRICE_CLOSE,4+cnt)/100);
where cnt is the loop counter.. this in essence means that we intend to train the network to just this sample.. why would we wanna do that?

3. ive modified the code to include the counter variable in indicator calculations.. i use the open-source fann library and its open-source gui fann-tool to train and test networks.. for 5000 data points, the mse doesnt seem to fall below 0.0045.. plotting the trained network's output against the training output shows the network hasnt generalized that well at all even at this mse.. the question.. what sort of mse should i be aiming for and which tools have you guys used to train the network?

thanks a million for your help in advance guys.. love this thread!!! :)

asim..
 
Hi All,

I would like to show you my own real trading result, implemented on the Indonesian stock exchange and use for long term trading strategy. (Sorry not forex trade or day trader).

All data are taken freely from my broker (or you can download also from yahoofinance), I implement all theoretical stuff that I learned using neural network. There are some reasons:
- I do not need to pay my data subscription (like CSI for VantagePoint), then I prefer to choose NN software that can work freely with different data format (txt, csv, xls, etc)
- I just need end of day data (no need a real time tick/minute for long term trade such as esignal or else).

So far my equity was growing more than to 20% profit and this is a real trading...not a joke..not a cheat...

Hope you can share your result as wells.

Arryex
www.ai-traders.org
 

Attachments

  • my stock data from yahoofinance.PNG
    my stock data from yahoofinance.PNG
    120.6 KB · Views: 1,342
  • my stock profit-loss calculation.PNG
    my stock profit-loss calculation.PNG
    139.6 KB · Views: 1,593
Top