Log in

Elliotware with Trading Solutions

image
 

 

Here I add another shot of the Elliotware approach. This time I used Trading solutions software. 

I use an expert for data link between MT4 and Trading Solutions software. The MT4 link updates the history and I use the authomatically generated csv to update the Trading Solutions history with every new bar. It is not really a data stream but it works.

The indicator allows me to export the data from the calculations made from MT4. And that is the big advantage. The big advantage is because I can export the SSA ep data from MT4 and use that data as an input for the Neural nets. 

In this case I used the SSA end - pointed and that data was used as an input for common indicators. For example you can use MACD from Trading solutions as an input but you can precise that instead of close you will use the SSA end pointed. And you have MACD with SSA end - pointed as an input. The most insane combinations are possible.

The other solution is to export the custom indicator from the MT4. In this case it is needed to update the i custom function in the data link indicator (i have SSA end pointed as i custom  function in the data link).  

 

I export data and make the training from exactly the beginning of this month. This is the Elliotware part of the things.

 

Comments

  • JohnLast 3690 days ago

    Here I would like to add the continuation of the model and to compare with another model that is using the same neural net model.

     

    image
    От 23 септември 2011

     

    image
    От 23 септември 2011

     

    The first model is the continuation of the old model. The model did perform very well. It was excellent (unfortunately I still hold two very small long positions :(, that I am thinking now how to cover).

    OK those models are default models. Really default, i mean they are shown as example models . The inputs are:

    Percent difference from median (period 10)

    Percent difference from median (period 20)

    Stochastic oscillator

    Vertical Horizontal Filter

    And those models are trying to model the optimum signal. Imagine the optimaum signal is something like the meta trade Zig Zag. And we try to use the inputs in order to model with neural net those Zig Zag signals.

    OK one of the models is using the default model as it is (the unsuccesful one). The other is using the SSA end - pointed instead of close price to calculate the other indicators: so we had a Percent difference from median

    The model with the SSA end - pointed performed much better than the model using close prices.

    However if you look with what happened today with the euro, it lost arround 400 pips in just two days. Wow. The loosing model gave correct signals if the trend was going to continue. And I used that kind of signals from another model. Well, I think that a neural net, you really cannot expect from it to predict changes in trend. Do you? I think the loosing model was correct but the assumption was that the trend was going to continue (In fact we still do not know there are many possibilities, for example broading triangle and resuming the up trend, it is really not sure yet).

    In fact I am not blaming the loosing model. I am surprised that the model with SSA performed well.

       

  • JohnLast 3689 days ago
    image
    От 23 септември 2011

     

    I have a buy signal now on 1 h chart using the Elliotware approach.

     

  • JohnLast 3687 days ago
    image
    От 23 септември 2011

    Ups I did not confugure well the screen shot.

  • JohnLast 3685 days ago
    image
    От 23 септември 2011

    Here I add another screen shot. As the market is not open the data link does not work. But according to the last data the model was perfect until now.

    Take care this week is going to be tough. I woud be curious to update the model to see if I have another buy, or the sell is still valid.

  • JohnLast 3684 days ago
    image
    От 23 септември 2011

    Here I add the last signal from the Elliotware sequence. If there is a break - out from the up side ther should not be training, but if it break down, maybe I should retrain from the beginning od the down trend.

  • JohnLast 3680 days ago
    image
    От 23 септември 2011

    I am quite surprised to tell you the truth. The system played quite well the last days.

  • JohnLast 3676 days ago
    image
    От 23 септември 2011

    Here I would like to add the results that are out of sample from the model. It looks quite good.

    It looks like the pictures above are not working, shit Google with its recent update og Picassa has created a complete mess. Google sucks regularly.

     

  • JohnLast 3676 days ago

    I can say that I am quite happy with this model with Trading Solutions. I even think that the use of Singular Spectrum Analysis combined with the common neural nets is a good combination.

    Here what is original is that I use the SSA in the computation of standard technical indicators. 

    -Percent difference from median lag 10

    -Percent difference form median lag 20

    -Stochastic oscilator

    -Vertical horizontal filter

    I model the optimum signal. (simething like a perfect zig zag in the training range) and the I try to model this perfect signal with my inputs.

    In fact this is the generic model in the software. I apologize because I had to make a benchmark between the original model with the close and the model with the SSA end-end pointed inputs.  

    In fact the preprocessing with SSA makes the hard job of preprocessing the information. Then you need to give that information to the neural net hoping that the market would maintain its characteristics meanwhile.

    And the Elliotware approach is trying with the human pattern recognition to watch for external characteristics in the data.

    And after all as we do not know how the training works and what it finds, a luck is needed because every training is unique and gives a unique solution. 

     

     

  • JohnLast 3663 days ago
    image
     

     

    This will be the last trade from this sequence. As it is time now to see the results from the Elliottware sequence (machine learning extentions of the Elliott Wave principle).

    The first shot shows the trades. Somewhere from the 15,16 of November the market started to get out of the possible Elliott Wave sequence as the market is clearly bullish.

    As you can see the training was done during an impulse sequence and we made a cross validation on a ranging phase just to make sure that

    the solution obtained during the impulse is valid during on the ranging phase.

    Thetesting was also on this ranging phase and the beginning of the new impulse.

    Those periods are clearly shown with the green vertical lines.

    As a result we had out of sample solution that holded quite well both in the subsequent impulse and ranging phases. 

    As the pattern exit the expected pattern the solution started to degrade rapidly. That is why I choose to stop this sequence now. I would stopped it even earlier but I wanted to go to the end and obtain the results.

     

     

  • JohnLast 3663 days ago

    Here I will post the statistics of the results. The pourcent of the profitable signals started to degrade rapidly when the bear trend was confirmed on the 18 th of November. I started with 80 % winners and that started to degrade to end up with 52.9 %. The strategy outperformed the buy and hold strategy and had very controlled risk. 

     

    image
     

     

    image
     

     

    image
     

     

    image
     
  • JohnLast 3646 days ago

    Here I would like to add another out of sample analysis using the same approach with Trading Solutions. However this time the training range is from 31 November in order to use the most recent patterns according both to the Elliotware approach and Market State Analysis. 

    However I would use a different system it is also one of the default systems however instead of market Close I use the Singulat Spectrum Analysis end - pointed imported from Metatrader. 

    The System uses 4 percent differences from the SSA ep with different periods.

    Percent difference From SSA ep with period 12

    Percent difference From SSA ep with period 78

    Percent difference From SSA ep with period 390

    Percent difference From SSA ep with period 390

    Here are the results:

     

    image
    От 23 септември 2011

     

    image
    От 23 септември 2011

     

    image
    От 23 септември 2011

     

    image
    От 23 септември 2011

     

    image
    От 23 септември 2011
  • JohnLast 3638 days ago
    image
    От 23 септември 2011

    I show how this works out of sample, however I see some displacement of the zones of the ranges, but it was me, however the signals are the same and are not displaced. 

    Now you can see  that it worked out of sample pretty well. I am glad. Right now for the Sell signal I would not say anything because we are at Xmass time so it is better not to trade. 

    I think I am going to continue to test this methodology of determination of the training ranges for expert systems and neural network systems. 

    This appears to be a different paradigme. A kind of meta knowledge about the systems. In that way traders can communicate and collaborate together without reveiling their proprietary knowledge and know how, because it is not that what really matters. 

    Actually I am working on market states identification. The idea is to detect a stable feature in the market. One of those features is the market state, the other is the switch from one market state to another regarding volatility. 

    The idea is that what really matters is to find a relatively stable characteristics (pattern but not necessarily a price pattern). The system comes after, in fact a system can be generated for minutes unless you know when and what to look for. 

    The questions are: 

    What conditions exactly? 

    What are their characteristics: volatility, some visual patterns, fractal dimension (I find it very useful)? 

    Are they recurrent? 

    And can you predict when those conditions will come again and how long they are going to stay? 

    Can we have a separate classification mechanism that is going to confirm objectively that we are in the right conditions? 

     

  • JohnLast 3631 days ago
    image
    От 23 септември 2011

     

    Here I add the last results with the test with Trading solutions. I used the same training range as for the out of sample neural net experiment with Neuroshell Day Trader. No really I am glad to share those results with you, I really like it. 

    Annualized return 48.10 %

    Percent winners 57.1 %

    Profit factor 2.94

    Sharpe (Annualized) 2.01 

    Maximum  drow down from investment -0.13 %

                                 from high equity -1.25 %

     

     

  • JohnLast 3631 days ago

    Those tests for the last screen shots worked almost like the perfect signal service LOL LOL LOL.