The tool supports interface of Foresight Analytics Platform 9 or earlier.
The Additional Parameters panel is displayed after selecting a method on the Parameters panel:
The set of available parameters depends on the type of configured model. The following model parameters can be set up:
In the Method drop-down list specify the missing data treatment method in variable values. The methods are identical to methods of the Missing Data Substitution model. Additional methods are also available:
None. Missing data is not treated.
Casewise. It is used by default. Empty values are excluded from the series. They are ignored in the calculation.
Custom. The option is available only for the Aggregation (Extended) model. The custom method of missing data treatment is set by using the Fore language (the property IMsCrossDimensionAggregationOptions.FillGapsUserMethod). If the custom method has been replaced with a standard one, and the changes have been saved, the user cannot select a custom method of missing data treatment in the drop-down list.
Features of missing data treatment in the Linear Regression (OLS Estimation) and Linear Regression (Instrumental Variables Estimation) models:
If a variable does not use the missing data treatment method defined for the model, the Missing Data Treatment group also includes the Apply To drop-down list. The list contains objects, for which a missing data treatment method can be set. The list contains:
Model. Default value. The missing data treatment method is applied to the output variable and to the factors and instrumental variables, for which the Apply Model's Method of Missing Data Treatment checkbox is selected.
Model factors or instrumental variables, for which the Apply Model's Method of Missing Data Treatment checkbox is deselected.
NOTE. The Apply Model's Method of Missing Data Treatment checkbox is displayed in the context menu of a factor or instrumental variable.
If several objects that have various previously defined different methods of missing data treatment are selected in the Apply To list, a confirmation is required to apply the same method to all of these objects.
Confidence limits relevance level is a probability that an unknown value falls into a confidence range. The confidence limits relevance is set by the 1-alpha value, where one of the standard significance levels (0.1, 0.05 or 0.01) is used as alpha. For example, for alpha=0.05 the confidence limit is 1-0.05=0.95, that corresponds to 95%. This value is used by default.
To set confidence limits relevance level for a forecast series, use the Confidence Limits Relevance box. Available values range: (0; 1).
On calculating confidence limits in the Linear Regression (OLS Estimation) and Linear Regression (Instrumental Variables Estimation) models, it is assumed by default that coefficients are calculated approximately, that is, the Take into Account the Uncertainty of the Coefficients checkbox is selected. To calculate confidence limits without taking into account that they are calculated approximately, deselect the Take into Account the Uncertainty of the Coefficients checkbox.
For the Linear Regression (OLS Estimation), Linear Regression (Instrumental Variables Estimation), ARIMA and Panel Data Model models, the Additional Parameters panel contains the Parameters of ARMA Estimation group:
The group is available if autoregression order and/or moving average are set for a model. Set the following parameters of ARMA estimation:
Calculation Accuracy. Select accuracy of model calculation. The minimum value is 0.00001. The default value is 0.0001.
Number of Iterations. Set the maximum number of iterations, during which the optimal problem solution is to be searched. If the iteration maximum is reached, but no optimal solution is found, the Errors panel shows the following message: "No satisfactory solution found for the specified number of iterations". In this case, increase the maximum number of iterations. When the number of iterations is large, the accuracy of calculation is the highest, but more time is spent.
Minimum value: 1. Default value: 500.
Estimation Method. Select the method to estimate coefficients of autoregression and seasonal average in the drop-down list.
Use BackCast. The parameter is available if the moving average order is set for a model. If the checkbox is selected, back-casting (building a line chart for the past period based on the trends and data of the recent past and the current period) is used to estimate the moving average coefficients.
NOTE. Only the Calculation Accuracy and the Number of Iterations parameters are available for the panel data model.
The Additional Parameters panel contains the Estimation Parameters group for the Non-Linear Regression (Non-Linear OLS Estimation) model:
Set model estimation parameters:
Calculation Accuracy. Specify accuracy of equation calculations. The minimum value is 0.00001. The default value is 0.0001.
Maximum Number of Iterations. Specify the number of iterations for model calculation. Minimum value: 1. Default value: 500.
Estimation Method. Use the drop-down list to select the estimation method for non-linear regression coefficients.
Use Analytical Derivatives. Selecting this checkbox increases calculation accuracy by calculating analytical derivatives for the non-linear regression equation.
Use Default Initial Approximations. The checkbox is selected by default, and default initial approximations of equation coefficients are used for calculation. If the checkbox is deselected, custom initial approximations are used. To view and edit initial approximations, use the Identified Equation panel.
See also: