Announcement

Collapse
No announcement yet.

AI4Forex - Martingale Grid self optimized with machine learning

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • AI4Forex - Martingale Grid self optimized with machine learning

    This is the "incubating" project based on the expert published in this topic: https://www.forexsignals.com/forum/f...st-ea-attached.

    The strategy is completely automated. It studies the last 3 months to get what I call the "DNA" of the instrument and uses machine learning to chose the best values for the next month.
    You can find all the backtest and a free to test expert in the topic I mentioned before.

    I started a test on a cent account with 1.000 $cent. The setting is extremely aggressive in order to achieve a double digits return every month.
    Max drawdown setted is all the equity, so.. no SL.
    This is the first public live test.
    If results will be good and I will see interest in this project I will start a Signal provider. If someone is interested in the project, like the expert and results and want to make a partnership, just sen me a PM.

    Myfxbook


  • #2
    Going deeper into the description of the strategy, I want to show what "DNA" means.
    I want that the grid could be able to open counter trend, not suffer to much if prices go against the entry and have a good exit point at the first retrace.
    Basically I need to know this 3 main parameters:
    DNA.png
    a) The best moment for counter trend entry. How much the price needs to fall before I enter the first Buy trade?
    b) How much "reasonably" price could go against me before the retrace?
    c) How big the retrace need to be in order to close the basket in profit?

    If we could guess these 3 numbers we will have the perfect trading system for all market conditions. We know that every market periods and instruments have different values, but they have... We will never see a market with only point A and B, without C, sure!

    So, the machine learning try to guess these 3 parapeters for the current market condition.

    Nice, but what happens if market change during our live session?
    Good question! I'm developing a more complex algorithm that could guess how much these parameters could change in a period of a month, in order to be prepered for these changes.
    In simple words:
    if volatility will increase at a record high, how much it will increase in a single month? Could it change from 100 pips per leg to 1000 pips per leg? Or it could increment maximum of 50%?

    Solving this problem will create the perfect grid.
    Stay tuned!
    Last edited by IA4Forex; 03-14-2018, 10:49 AM.

    Comment


    • #3
      Originally posted by RenkoGuy

      On what time frame do i trade ?
      Can you remove the Lotevery ? its anoying it doesnt take any trades
      On the MT5 version
      The time frame is not important, it's setted by the logic.
      In the first post I updated the MT5 version. Now if you set LotEvery to 0 it will use base lot

      Comment


      • #4
        Originally posted by RenkoGuy
        Hi your link to the Ea isnt working
        It has been stopped by the Spam filter...I already wrote to the admin. For now I'm posting here the mt5 version.
        Attached Files

        Comment


        • #5
          I started the main strategy with 3 different sets, 2 on EURUSD and 1 on NZDUSD.
          I won't stop the EA during this week news to stress it, but I have to say that the broker is really really poor. Ok, it's a cent account, but execution is more than slow...
          Anyway, go on and see.
          1903.png

          Comment


          • #6
            Quick Update. The monthly target has been reached after 8 trading days.
            10% profit with 5% drawdown. The next week the grid will go on with the same neural network.
            During the next week end I will train the new settings based on the last 3 months data (March included).
            DD will increase for sure, settings are extremely aggressive.


            AI4Forex.png

            Comment


            • #7
              At the moment the market conditions are perfect for a grid/martingale system. It'll be interesting to see it perform in a trend market.

              On the latest bigger move (20.3) i didn't see anything that would make this system special to be honest. Looking at the trades it was the most basic grid strategy.

              But let's see how it performs over the year.

              Comment


              • #8
                Originally posted by Amadorian View Post
                At the moment the market conditions are perfect for a grid/martingale system. It'll be interesting to see it perform in a trend market.

                On the latest bigger move (20.3) i didn't see anything that would make this system special to be honest. Looking at the trades it was the most basic grid strategy.

                But let's see how it performs over the year.
                Hi Amadorian, you are right, it's a simple grid strategy as we see many, but:
                - Step is dinamically calculated by machine learning (ML)
                - Entry point is decided by ML
                - Lot factor given by ML
                - TP given by ML

                I've seen many strategy during 10 years, some of them were very impressive, like for example Forex Growth Bot, based on solid Tecnical base. But when market change all of them have failed.
                That's why I use a simple idea of a grid, but all main parameters are setted by a neural network.
                I still have to understand what period of training is the best for the neural network.
                I mean, If I give 18 years to the algo, I have a grid that overcome all (more or less) trading market, but gives poor return like this: https://www.forexsignals.com/forum/f...7159#post97159
                But if I gave to the net the last 6 months, train it for 1 forwqard month, I have a great chance that the grid will works for 2 or 3 weeks with double digits return.

                Keep in mind that this project is a work in progress, so I adjust the strategy over time.

                If you want I could train a particular period, difficult for grid, and try to give the six month before to the machine learning and see what would have been the results after the training.

                Comment


                • #9
                  Why a grid? Because these kind of systems are very active, dozen of baskets in a week and positions are hold for few hours.
                  In this way it's easy to understand if the current settings are good or if the market conditions are changed.
                  Let's say that we trained the grid in the last six month and max DD was 8%. We could run for a couple of week with stop loss of 10%. If market change just in these 2 weeks, we are stopped out with "small" loss. Usually market conditions change few weeks per years, let's say one or 2, so not a great issue.

                  The real problem is the computation time. With my PC I can train only one pair, but it would be interesting to test all majors at least.

                  Comment


                  • #10
                    I need to stop this test cause the broker has a slippage of 1.5 pips per trade. The real cost on the account is equal to the profit, so, with a decent broker the 3 systems would have profit 40% instead os 20%.
                    I won't restart with other broker because I don't see interest here.
                    Thank you


                    report.png

                    Comment


                    • #11
                      Ciao Walter, hai provato a far funzionare il tuo EA su di un buon broker con spread bassi come ICMarkets ?

                      Comment


                      • #12
                        Originally posted by laracroft View Post
                        Ciao Walter, hai provato a far funzionare il tuo EA su di un buon broker con spread bassi come ICMarkets ?
                        Ciao, personally I use Interactive broker, the best broker for reputation, execution, liquidity and number of instruments.
                        They don't have MT4, but it's not a problem for me because I code in java and pyton, so I connect to TWS with API.

                        Comment

                        Working...
                        X