The General tab provides the most basic information about a forecast step.
Name
You cannot change the name after creating the step.
ID
The full logical name.
This information appears on the property sheet after the step has been created; it does not appear on the initial dialog box.
Short Label
A short descriptive name for display, usually in mixed case.
Long Label
A long descriptive name for display, usually in mixed case.
Description
Additional descriptive text, which can include mixed case and spaces.
Cube
The cube containing both the source and the target measures.
Source Measure
The measure containing historical data values, which will be used to generate the forecast.
Target Measure
The measure that will store the forecast values.
The target measure and the source measure can be the same. Over time, however, the actual values will overwrite the forecast values, so that you will be unable to monitor the accuracy of the forecast. By defining a different measure in the cube for forecasting, you can compare the forecast and actual values.
Time Dimension
The name of the time dimension whose periods will be forecast. You only need to choose a time dimension if the source measure has more than one, such as separate dimensions for months and years.
Forecast Method
The statistical method used to generate the forecast. Select a method to display its description in the Forecast Method Description box.
The forecasting methods can be grouped into the following categories:
Automatic: The forecast engine identifies the best fit by quickly testing each statistical method against the historical data. It selects the method that would have generated the most accurate forecast in the past.
Linear Regression: Attempts to fit the historical data to a straight line, and extends that line into future time periods for the forecast. All data points are given equal weight. This method identifies steady long-term trends in the data.
Nonlinear Regression: Attempts to fit the historical data to a curved line, and extends that line into future time periods for the forecast. All data points are given equal weight. The curved lines are defined by mathematical equations. You can choose from the following types of curves:
Polynomial: Fits data that fluctuates with a rise and a drop.
Exponential: Fits data points that rise or drop at an increasingly faster rate.
Logarithmic: Fits data points that rise or drop quickly and then level off.
Asymptotic: Fits data points that rise or drop until they approach a fixed value and then level off.
Exponential Asymptotic: Fits data points that rise or drop at an increasingly faster rate until they approach a fixed value and then level off.
Exponential Smoothing: Calculates a single smoothing parameter for all forecast periods. The prior period has the most weight and each period prior to it has comparatively less weight. The decline in weight is expressed mathematically as an exponential function. You can choose from the following methods of exponential smoothing:
Single Exponential Smoothing: Does not adjust for trend or for seasonal variance.
Double Exponential Smoothing: Adjusts for trend.
Important: If you choose a method other than Automatic, be sure to set Forecast Approach and Data Filters on the Advanced Parameters tab.
Number of Forecast Periods
The number of time periods for which you want to forecast data. These periods must already be defined in the time dimension.
Forecast Method Description
Displays a description of the selected forecast method.
Equations for Forecasting Methods
Copyright © 2003, 2007, Oracle. All rights reserved.