Signal Postprocessing Settings Dialog

The Signal Postprocessing Values Dialog allows you to view and modify the settings associated with postprocessing signal values that are produced by predictions.

Ä    Note: Fields that are currently being used by the Solution Service cannot be modified.

&  For help with predictions, see Predicting and Modeling Financial Data.

Understanding the Signal Postprocessing

Predicted entry/exit signals often hover around a single value. This is because when neural networks lack adequate information to make perfect predictions, they tend towards the mean of the desired outputs. This does not mean that the prediction is not valid. However, in the case of entry/exit signals, it may result in an output that remains too close to the mean to cross over the thresholds that differentiate entry/exit signals.

To adjust the predicted entry/exit signals to be usable as signals in TradingSolutions, postprocessing is applied to the values output from the network. This basic postprocessing uses values to adjust the outputs by establishing thresholds within the range of outputs. These values are normally set using a brief genetic optimization stage after the training.

Selecting the Type of Postprocessing

Two different types of signal postprocessing are available. In most cases, the default setting is used.

¨    Type of Postprocessing

·      Adjust Thresholds (default)

This is the most common option. Internally, TradingSolutions uses constant thresholds associated with entry and exit signals. This option allows you to set different thresholds to use for the predicted signal. These field-specific thresholds are then used to amplify the predicted signal to the appropriate levels.

·      Offset and Multiply

This option is a less complex type of postprocessing that is typically used when the postprocessing values will be hand-modified. An offset is added to the predicted signal. Then, the sum is multiplied by a scaling factor. Since the postprocessing values affect all of the predicted signals in the same way, they are typically easier to modify by hand than the values associated with adjusting the thresholds.

·      None

This option is typically used when you would like to see the raw output of the prediction. It is also useful when you would like to apply your own postprocessing algorithm.

Ä    Note: For more information on how these values are used, see the section on Current Postprocessing Settings, below.

Understanding the Current Postprocessing Settings

The current postprocessing settings that are displayed are dependent on the type of postprocessing that is currently selected for this field.

·      Adjust Thresholds Postprocessing

These values represent the thresholds used against the values produced by the neural network.

p    Enter Long

Outputs greater than or equal to this value are adjusted to be 0.5 or greater. Scaling of larger values is based on the distance between the Enter Long and Enter Short thresholds, or Enter Long and zero if they are equal.

p    Exit Short

Outputs greater than or equal to this value, but less than the Enter Long threshold are adjusted to be in the range from 0.2 to less than 0.5. Scaling is based on the distance between the Exit Short and Enter Long thresholds.

·      (Hold)

Outputs between the Exit Short and Exit Long thresholds are adjusted to be in the range between -0.2 and 0.2. Scaling is based on the distance between the Exit Short and Exit Long thresholds.

p    Exit Long

Outputs less than or equal to this value, but greater than the Enter Short threshold are adjusted to be in the range from -0.2 to greater than -0.5. Scaling is based on the distance between the Exit Long and Enter Short thresholds.

p    Enter Short

Outputs less than or equal to this value are adjusted to be -0.5 or less. Scaling of lesser values is based on the distance between the Enter Long and Enter Short thresholds, or Enter Short and zero if they are equal.

·      Offset and Multiply Postprocessing

These values represent the terms that are used when postprocessing the outputs of the neural network. They are performed in the order offset, then multiplication.

p    Offset

This value is the amount added to the output of the neural network before it is multiplied.

p    Multiplier

This value is the amount by which the output is multiplied after the offset has been added.

Configuring Postprocessing Optimization

The postprocessing settings are normally set during a brief genetic optimization phase following training. This optimization is performed by simulating trading of signals processed with potential values until the best solution is found.

The following settings are available for modifying that optimization.

þ    Optimize postprocessing settings.

This option specifies that signal postprocessing should be optimized using genetic optimization after each time the prediction is trained. If this value is not selected, TradingSolutions will select basic values after the first training based on the range of output values.

þ    Optimize trading style settings.

This option specifies that selected trading style settings should be optimized along with the postprocessing settings.

Ä    Note: This option is only available if the trading style is being overridden for the current field. For more information, see the help for the Modify Field Dialog: Trading Style page.

þ    Re-optimize postprocessing settings on save.

This option indicates that these settings should be re-optimized when these changes to the field are saved. This is useful if you change the trading style or other optimization settings.

þ    Restrict optimization of signal based on conditional inputs.

This option indicates that the signal should only be optimized over ranges where any conditional inputs to the prediction are true. If this option is selected and one or more conditional inputs are false, the value of the signal is set so that any current positions are exited.

Ä    Note: The entire optimization range will still be analyzed. Therefore, this setting may cause settings such as the minimum number of trades in the Trading Style to be have a different effect, depending on how often the conditional inputs are false in this range.

p    Optimization Settings…

This button displays the Postprocessing Optimization Settings Dialog, which allows you to modify options for which there are global defaults.

What Do I Do Next?

When you are finished modifying the settings, press the OK button to save any changes. If you would like to exit this dialog without saving your changes, press the Cancel button.

How Did I Get Here?

The Signal Postprocessing Settings Dialog appears if you press the Signal Postprocessing Settings… button on the Modify Field Dialog: Training Analysis: Overview page. It also appears if you press the Values… button on the Modify Field Dialog: Desired Output page when the desired output is a signal and the definition is for a data series. Postprocessing values for group-level predictions are not available from the definition.