Adaptive TurboProp2 (AT2) is the same basic neural network algorithm that is used in the NeuroShell Trader, Trader Professional, and NeuroShell DayTrader Professional Prediction Wizard. However, AT2 can retrain itself as frequently as every new bar or after a number of bars which you can specify. The training set is always the most recent bars. AT2 does not make predictions on the training set bars, and its predictions are always "out of sample" (exception: when you specify that you are predicting zero bars ahead). Therefore, there are no walk forward periods to worry about, since the net is essentially walking forward each day (or each bar in the case of the NeuroShell DayTrader Pro).

An AT2 net may be used as an indicator, as an input to a regular TurboProp2 net in the Prediction Wizard, or it may be used directly in a Trading Strategy. The output is a signal, meaning that the prediction is shown on the current bar, even if it is a prediction for 10 bars from now.

AT2 doesn't require the optimizer in the Professional versions of the Trader because the parameters are so easy to set. However, when used with the NeuroShell Trader Professional or NeuroShell DayTrader Professional, you can optimize the following parameters:

1. trainbars - the training set size (number of recent bars on which to build the model), up to a maximum of 10,000.

2. aheadbars - the lookahead period (the number of bars into the future the prediction is to be made), i.e., the number of bars ahead that the actual signal is predicted. Be careful not to allow 0 in the range, because then you are predicting the current bar, not the future.

3. actual - the parameters of any indicator used as an output.

4. hiddens - the number of hidden neurons to use when making a network (Turboprop2 in the prediction wizard automatically decides how many to use, but AT2 lets you or the optimizer decide). You can use up to a maximum of 20 hidden neurons, more than enough for this adaptive version.

5. retrain - this parameter specifies in bars how often the net is trained. If retrain = 1, a new net is created for each new bar encountered. If retrain = 10, the net is retrained every 10 bars.

6. input1, input2, etc. - these parameters are the network inputs. You can optimize all parameters of indicators used as these inputs.

Available Configurations

AT2 nets allow 2 to 14 inputs. The names of the indicators are correspondingly Tprop2, Tprop3, …, Tprop14.

Adaptive TurboProp2 requires release 3.2 or better of the NeuroShell Trader, NeuroShell Trader Professional, or NeuroShell DayTrader Professional.

For more details, please view the product manual for this add-on