Forex Neural Expert Advisor Neural-AutoTraining

Forex Neural Expert Advisor Neural-AutoTraining - Summary

Expert Advisor based on multi- layer neural network (BackPropagation Neural Network ) with fully automated neural net training on the end of the week. Forex Neural Net will analyse all losing orders and inaccurate enters/exits for previous week and train based on new information. Expert Advisor will start trading on Monday with new re-trained neural net. This cycle will start again next weekend.

Forex Neural Expert Advisor Neural - Description

A core of the forex neural expert advisor is a scalping module, which trades during periods of quiet market. The sell signal is generated at achievement of maximum; the buy signal is generated at achievement of minimum.

The signal is estimated by the neural network. A complex BackPropagation neural network architecture is used in the advisor. In our opinion, such network topology is perfectly suitable for prediction of time series behaviour. Imperfection of neural networks as tools for prediction of financial markets is that accuracy of predictions decreases in time. Thereby, we’ve developed «the auto-training module», which allows to retrain a neural network automatically at a certain time. 

User can specify the training time. For example if you want to train neural net every Sunday at 1pm on historical data for 6 month, you should specify:

OptimEveryTime="13.00";   (0-24 hours)
extern string OptimEveryWeek="7";   (1-Monday, ....7 -Sunday)
extern int OptimPeriod=180; (days)
 

The training mode of the network is activated during weekend. After training the module deletes old files of the neural network and writes down the new ones. The advisor starts to trade with a new network on Monday. The advisor works in a completely automatic mode and does not require any actions to be done by the trader. 

BackPropagation Neural Network - short explanation

FeedForward BackPropagation architecture was created in the beginning of 1970s by several independent authors: Werbor, Parker, Rumelhart, Hinton and Williams. Now ВackРropagation paradigm is the most popular, effective and easy learning model for complex multilayer networks. It is used in different types of applications and has generated the wide class of neural networks with different structures and training methods.
A typical BackPropagation networks has an input layer, an output layer, and at least one hidden layer. Theoretically, there are no restrictions concerning the number of hidden layers, but practically only one or two are used. 

 

Neurons are organized in a level-by-level structure with a direct signal transmission. Every neuron of the networks produces the weighed sum of its inputs, runs this value through transfer function and delivers an output value. The network can model function of practically any complexity, and the number of layers and the number of neurons in each layer determine complexity of function. Determination of the number of intermediate layers and the number of neurons in them is important at network modeling. 

Download Forex Neural Expert Advisor Neural-AutoTraining HandBook

Forex Neural Expert Advisor Neural-AutoTraining Back tests:

Portfolio (EURUSD + USDJPY + EURAUD ) Back Test

EURUSD Back Test

USDJPY Back Test

EURAUD Back test