We’ve used neural networks and applied them to trading Forex successfully in the past and decided to translate that method into a Metatrader system. It is widely known that the larget trading firms and hedge funds use sophisticated artificial intelligence and nueral network systems to profit from the financial markets with staggering accuracy. We thought, why can’t that power also be available to us - the small money investors?
So I took a break from all my other activities and worked hard with Metaneural to develop this system, which I believe to be the only REAL neural network EA. In fact, it doesn’t even have to be an EA, the code can be written in C++ to work exactly the same way in tradestation, esignal, neuroshell, or any platform that allows DLL importing and data gathering, because the neural network creation happens in Neurosolutions.
HOW IT WORKS
Data Colletion
The first step in creating an artifical neural network brain is to gather the data around which the structure of the brain will be formed. Since we are trying to create a brain that will know how to trade the markets we must gather market data.
However, we cannot simply collect a mass of data and dump it into our neural engine to create the structure of our brain. We must gather the data in the format which we want the brain to process that data and eventually the same format we want it to create output in. In other words, we’re not only telling our brain WHAT to think, by giving it raw data, but we must tell it HOW to think, by formulating that raw data into an intelligable configuration.
In this case, our intelligible configuration is patterns. We gather data in segments, each segment consists of a number of bars set by the trader in our proprietary collection indicator which comes with all of our packages. That grouping of bars is collected in relation to the next bar that comes after the grouping - we will call this the future bar. When we’re collecting market data the future bar is known, because it is all historical data, it is the next bar after the grouping. The idea is that the neural network brain will find complex patterns in the bar grouping and use the information collected, including the ‘next bar’ after the grouping, to determine which complex patterns preceed the result of the next bar. During actual trading that result will be the future bar which in effect makes it possible to know with a high degree of accuracy the direction of the market before it happens.
The collected data is extracted into a spreadsheet which displays price data as open, high, low, close (OHLC). The OHLC of each bar is collected separately and placed in its own column. In the example above each row represents 3 bars in total. Therefore, the columns represent hundreds or thousands of bars collected going back into history.In addition to OHLC you can also collect the values from almost any indicator you select, which will essentially give that indicator the ability to ‘think’ based on changing market conditions and predict the next value.
Neural Network Building and Training
Now that we have our collected data, extracted into a spreadsheet file in an intelligible configuration, we can load it into our neural network engine which will create the structure of the artifical brain, train it, and test its accuracy before saving the structure.
Once the collected data is imported into the network building program you are given the choice to select which bits of data you want to use to build your brain. This is an important feature because it enables the user to create many different strategies based on whichever piece of data is deemed necessary. What we’re essentially doing in this step is determining what the engine will use to create the complex patterns mentioned earlier, which will ultimately decided the projection ability of the neural network EA.
For example, say you wanted to tell the neural network to only look for patterns in the open prices of bars in relation to the indicator values from your favorite indicator. You would then select your indicator in the collector and choose only the open and data inputs in the building software depicted above. You can also select all the inputs, except for the output1 column, which signifies your output value - selecting all inputs will create the most complex learning pattern possible and thereby allow your brain to respond to many different scenarios.
Once the desired inputs and outputs are selected the software will create the structure of your neural network brain and you can begin to train it. A portion of the collected data is set aside and used to train and test the accuracy of your artificial brain, you will see the desired output begin to conform to the testing data as it ‘learns’. Once this process is complete you will be able to export the structured artificial brain in the form of a DLL which will be used by the MetaNeural EA.
EA Implementation
Once the brain is built, trained, tested, and exported as a DLL you can begin trading with an automated neural network brain that will see complex patterns that are impossible for a human to achieve.
EA
I’ve attached the EA to this post with the supporting files. This EA is protected, to get a unique ID to run it please contact me.
[MetaNeural EA_v218-rev2.zip](http://wikisend.com/download/353030/MetaNeural EA_v218-rev2.zip)
