Chapter Contents |
Previous |
Next |

Window Reference |

Use the Forecast Combination Model Specification window to produce forecasts by averaging the forecasts of two or more forecasting models. The specified combination of models is added to the model list for the series. Access this window from the Develop Models window whenever two or more models have been fit to the current series. It is invoked by selecting Combine Forecasts from the Fit Model submenu of the Edit pull-down, or from the pop-up menu which appears when you click an empty part of the model table.

`Series`-
is the name and variable label of the current series.
`Model`-
is a descriptive label for the model that you specify.
You can type a label in this field or allow the system to provide a label.
If you leave the label blank,
a label is generated automatically based on the options you specify.
`Weight`-
is a column of the forecasting model table that contains the
weight values for each model.
The forecasts for the combined model are computed as a weighted average
of the predictions from the models in the table using these weights.
Models with missing weight values are not included in the forecast combination.
You can type weight values in these fields or you can use other
features of the window to set the weights.
`Model Description`-
is a column of the forecasting model table that contains the
descriptive labels of the forecasting models fit
to the current series that are available for combination.
`Root Mean Square Error`(or other statistic name) button-
is the button above the right side of the table. It displays the
name of the current model selection criterion: A
statistic that measures how well each model in the table
fits the values of the current series for observations
within the evaluation range.
Clicking this button brings up the
`Model Selection Criterion`window to enable you to select a different statistic. `Normalize Weights`button-
replaces each nonmissing value in the Weights column with
the current value divided by the sum of the weights.
The resulting weights are proportional to original weights and sum to 1.
`Fit Regression Weights`button-
computes weight values for the models in the table by regressing the series
on the predictions from the models.
The values in the Weights column are replaced by
the estimated coefficients produced by this linear regression.
If some weight values are nonmissing and some are missing,
only models with nonmissing weight values are included in the regression.
If all weights are missing, all models are used.
`OK`-
closes the Forecast Combination Model Specification window and fits the model.
`Cancel`-
closes the Forecast Combination Model Specification window without fitting the model.
Any options you specified are lost.
`Reset`-
resets all options to their initial values upon entry to the
Forecast Combination Model Specification window.
This may be useful when editing an existing model specification;
otherwise, Reset has the same function as Clear.
`Clear`- resets all options to their default values.

The newly selected model is given a weight equal to the average weight of the previously selected models, and all the nonmissing weights are normalized to sum to 1. When you use the mouse to remove a model from the combination, the weight of the deselected model is set to missing and the remaining nonmissing weights are normalized to sum to 1.

Chapter Contents |
Previous |
Next |
Top |

Copyright © 1999 by SAS Institute Inc., Cary, NC, USA. All rights reserved.