Professional Documents
Culture Documents
Impressum
Terms of use
Home
NeoTradEx
About us
Buy online
FAQ
User's Guide
Quick start
To start using the software using the preset parameters and prepared neural networks do the following. To install the software you must have Administrator Rights. Install the package, providing your serial number. When asked by the system confirm your Administrator Rights to continue the installation. During the installation you have to provide the Metatrader 4 software installation folder. Upon the installation procedure you should be able to see the expert NeoTradeEx in the list of expert advisors in your Metatrader 4 terminal. Remark: under Windows systems b efore than Windows Vista one must run the installation under account with Administrator Rights, since those Windows systems do not give an option to provide administrator password during the installation process. After the installation is complete you should enable Experts in your Metatrader software. Please go to Service Settings Experts and check the check-boxes Enable Expert Advisors , Allow Live trading and Allow DDL imports, uncheck Ask manual confirmation , uncheck Confirm DLL function calls (see picture please).
After that choose the time-frame and currency pair you want to trade. Drag the expert advisor NeoTradexEx on the selected currency pair. Load the corresponding parameter file (e.g. "eurusd_h1.set" file for EUR USD trading pair for the one hour "H1" time frame), which you can find in the corresponding folder (usually C:\NeoTradEx\PresetFiles).
NeoTradEx USERS GUIDE (folder C:\NeoTradEx): NeoTradEx.chm Prepared Neural Networks and Preset Files may be found in folders C:\NeoTradEx\NeuralNetworks and C:\NeoTradEx\PresetFiles respectively.
Theoretical background
NeoTradEx package uses neural network ideas and digital filters approach to smoothing of the financial time series. The emphasis is on the using the most simple and reliable techniques such as multi-layer back-propagation neural network. Neural network approach has been used in financial series prediction since decades, and has proven itself to be relevant and consistent. Neural network approach is based on using a system of inter-connected elements (neurons) having non-linear properties. It is proven that a superposition of a large number of such basic elements is able to reproduce any correspondence between input and output data.
A basic neural network is organized in "layers" (see picture please). As it can be easily seen network layers can be split into three distinctive classes: Input layer, Hidden layers, and Output layer. Input data is applied at the input layer. The neurons of the input layer carry out the initial processing of the data. Output of the input layer is connected to the hidden layer which disentangles complicated dependencies in the data. After that the output layer produces the result of the neural network processing. Depending on the nature of the output data multi-layer networks can be divided in pattern recognition and regression networks. We use pattern recognition networks which are meant to predict the current market trend direction. The gist of the learning routine consists in the following. Using historical data, the raw time series are fed into the network. As the output of the neural network the local trend direction is taken, the trend direction is determined using a so-called ideal filter. Using the standard back-propagation technique, the network learns the connection between the input and the output data. After the network has been taught it can be used to predict the market trend for the new data, and thus can be used for trading and market analysis. Training procedure has to be launched by the user of the software. Also the user can assess the VALIDATION on the validation data. The learning quality is controlled by the Mean Square Error (MSE) which can also be controlled by the user. Raw financial time series exhibit a high level of non-regular behaviour. But the neural networks are able to extract signal from noise, and the smoothing of the input data is carried out by the networks itself. As the output for the network learning we use the current market trend. During learning procedure we determine it by using the historical data in order to obtain the real (not predicted) market trend. This approach is based on using so-called non-causal (ideal) filters, which are built using digital filters, suppressing noise, and providing for zero delay in market trend evaluation. This type of filters can be used only during learning procedure, due to its non-causality.
Installation
To install the software you must have Administrator Rights. Install the package, providing your serial number. When asked by the system confirm your Administrator Rights to continue the installation. During the installation you have to provide the Metatrader 4 software installation folder. After that, the installer will put all the necessary files in corresponding folders. The already prepared neural networks ane preset files containing in installation package as well as this USER'S GUIDE you may find in the corresponding folder (usually C:\NeoTradEx). Upon the installation procedure you should be able to see the expert NeoTradeEx in the list of expert advisors in your Metatrader 4 terminal. Remark: under Windows systems b efore than Windows Vista one must run the installation under account with Administrator Rights, since those Windows systems do not give an option to provide administrator password during the installation process. After the installation is complete you should enable Experts in your Metatrader software. Please go to Service Settings Experts and check the check-boxes "Enable Expert Advisors" , "Allow Live trading" and "Allow DDL imports", uncheck "Ask
manual confirmation ", uncheck " Confirm DLL function calls" (see please picture here).
Fixed: true to trade with a fixed lot, false to trade with a variable lot which is calculated as a fraction of the Free margin, capital available for trading. Lots: the value of fixed lot. Relevant ONLY in case Fixed=true Capital: the fraction of the Free margin, capital available for trading, to be a value below 1 (values around 0.1 are recommended). Relevant ONLY in case Fixed=false MaxOrders: the maximum number of orders your expert can open simultaneously. It is recommended to set at 1. _____________________________________________________________________________________________________________________
Trading mode
After the installation you already have prepared Neural Networks supplied with NeoTradEx software. You can use them for prediction and actual trading. Choose the corresponding parameter file for your time-frame and currency pair. Before applying the expert to real trading you can test it in the Strategy Tester. You may also decide to train your own networks. This procedure is described in the next Chapter and addressed to advanced users. It is recommended to test these Neural Networks using the data which was not used during the learning procedure. To test that the learning procedure was successful you can also try to run the expert using the learning data. The neural network produces prediction of market trend, a digit between -1 and 1. -1 signifies a strong prediction of negative trend, +1 strong prediction of the positive trend. It is reasonable to use the same values of Stop Loss, Take Profit and Trailing Stop parameters which were included in the preset parameters files provided in the installation package. Please note that the parameters of Stop Loss , Trailing Stop and Take Profit are not in price points but in time steps, and the actual value of them is calculated based on the volatility (the average price movement). For example, if we set the stop loss parameter at H1 time-frame to be 24, it means that we use the daily average price movement to estimate the optimal stop loss position. The same applies for Trailing Stops, if set its parameter to be 1 it means that we use an average price movement in one hour as the measure of minimal price movement for which it makes sense to set TrailngStop. The parameter for the Stop Loss is called VolatilityStop, for the Trailing Stop is VolatiliyTrail and for the Take Profit is VolatilityTake In all three strategies you can also choose different level of network signal for opening the positions. The position will be opened if the absolute value of a signal is larger than variable DeltaOpen. The position will be closed if absolute value of a signal will be larger than variable DeltaStop. If DeltaStop is larger than 1, then positions are NEVER closed using the signals. You can also try to optimize these parameters using the validation data. The remaining parameters in the properties of the trading expert control the money management. You can choose to trade by a fixed lot. In this case you choose Fixed parameter to be True, and choose the fixed lot value in the Lots variable. You can also choose to trade with a variable lot which is calculated as a fraction of the Free margin, capital available for trading. In this case you should set Fixed to be False, and Capital variable to be a value below 1 (values around 0.1 are recommended). Parameter MaxOrders determines the maximum number of orders your expert can open simultaneously. It is recommended to set at 1. If you prefer to trade manually, you can still use the neural prediction by using the indicator NeoPrediction (see chapter Indicators). It has values between -1 and 1, negative values denoting the falling trend and positive values denoting the rising trend.
Indicators
The package NeoTradEx contains also three user indicators: NeoVolatility NeoTrend NeoPrediction The NeoVolatility indicator shows the average price movement, i.e. represents the characteristic size of time series fluctuations. This indicator is used in evaluation of Stop Loss, Trailing Stop and Take Profit (see chapter Trading Mode). Mathematical definition: Sum( |Price(i)-Price(i+Step)|, i=1...Length ) / Length Parameters of the indicator: Step: the number of events (bars on chart) the difference in price (i.e. movement of price) is calculated for. For example for the H1 timeframe, if Step=2, then we want to evaluate the average price movement for 2 hours, if Step=24, then we consider the price change during one trading day Length: the length for the averaging. For example, if one has 500 then we average the price movement over 500 events (approximately 1 month for H1 timeframe) During the trading by expert advisor NeoTradEx the order parameters Stop Loss, Trailing Stop and Take Profit are calculated with Length=500 and with Step set up as VolatilityStop, VolatilityTrail and VolatilityTake parameters respectively. The NeoTrend indicator shows us an output of a digital filter applied to the price time series. The output of the indicator is used as an input during the process of Neural Networks training. The indicator has no external parameters. The NeoPrediction indicator shows the trading signals produced by neural networks, i.e. shows the neural network prediction. The neural networks produce prediction of market trend, a number between -1 and +1. Negative values denoting the falling trend and positive values denoting the rising trend. The closer the signal to +1 (-1) the s tronger is a neural networks prediction of rising (falling) trend on a market. If preferred one may use the indicator to trade manually. Parameters of the indicator:
Verbosity: 0 is less, 1 is more output into journals and log files. Depth: should correspond the Neural Network file names used for prediction, i.e. if we have EURUSD-xx-xx-xx-xx-x-20000-0-xx-xx.x, then Depth should be 20000 to let the expert to choose correct Neural Networks files for predictions. ValidationLength: should correspond to another number from the mask EURUSD-xx-xx-xx-xx-x-20000-0-xx-xx.x, i.e. 0 to let the expert to choose correct Neural Networks files for prediction. 0 here is rather an exception to mark the network files provided with package from NeoTradEx.com. If one wants to train the network by himself, this number would be rather not 0 in order to evaluate the predictability on validation events. For example one uses ValidationLength 200, then the neural network files will be something like EURUSD-xx-xx-xx-xx-x-xxxxx-200-xx-xx.x. IndicatorDepth: the number of bars on chart the line for the prediction will be drawn. One should not take this number to high since it requires some RAM and CPU capacity of your PC. If you experience problems with the representation of the indicator on chart, then decrease IndicatorDepth to 100. After successful start of the indicator NeoPrediction, you should observe messages, similar to the one shown below, in the expert tab of your terminal, otherwise something is not correct:
Note: if you do not need indicators, we recommend to remove these from the chart to increase the performance of your PC!