The Advanced Settings tab provides numerous parameters that adjust the way a customized forecast is run. For the Automatic method, the forecasting engine selects the optimal parameter settings when it runs the forecast. For all other methods, the forecasting engine uses the default settings unless you provide different ones.
Forecast Parameter
Lists the various settings that you can modify. Select a setting to display its description in the Forecast Parameter Description box.
The parameters can be grouped into the following categories:
Not all parameters are available for all forecasting methods.
Setup Parameters: These parameters provide the forecasting engine with basic information about how you want it to approach a forecast. Always set the Forecast Approach and Data Filter parameters when using a specific forecasting method.
Forecast Approach: Specifies the approach that the forecasting engine takes in applying the Expert System. Automatic gives full control to the Expert System. Manual enables you to choose a method and parameter settings that are appropriate for the historical data.
Data Filter: Identifies a basic characteristic of the data for data filtering. Set this parameter whenever you are using a method other than Automatic.
Verification Window Size: The verification window is specified as a fraction of the total number of historical periods. The Expert System uses the verification window to determine the best method and parameter settings.
General Parameters: Use these parameters with any specific forecasting method.
Allocate Last Cycle: Controls whether the last cycle is calculated using allocation or forecasting. Allocation may reduce the risk of overadjustment for trend or seasonality. This parameter depends on the setting for Periodicity.
Periodicity: The number of periods in a single cycle or the number of periods in each set of nested cycles. The default value of 1 does not group the periods at all, so each period is logically independent.
For example, if you are using Month as the base level for the forecast, and the time hierarchy has levels for Month, Quarter, and Year, then the cycles are 12 months in a year and 3 months in a quarter. For a single cycle, enter the number of periods. For nested cycles, list the cycles in parentheses from the most aggregate to the least aggregate, separated by commas, such as (12,3).
Boundary Maximum and Boundary Minimum: Forecasting boundaries constrain the forecasting engine from occasionally generating unreasonably high or low values. The upper boundary is calculated by multiplying Boundary Maximum by the largest value in the historical series. The lower boundary is calculated by multiplying Boundary Minimum by the smallest value in the historical series.
For example, if the Boundary Maximum parameter is 100.0 and the largest historical value 5,600, then no forecast value can be greater than 560,000. If the Boundary Minimum parameter is 0.5 and the smallest historical value 300, then no forecast value can be less than 150.
Moving Total Decay Maximum and Moving Total Decay Minimum: The maximum value of a decay constant that is inversely related to noise, random deviation, and stability in the history of intermittent data. Set this value higher when the history is evolving rapidly from one cycle to the next or when the noise level is low. This parameter is used only with the Intermittent Data filter. The difference between the maximum and the minimum must be evenly divisible by 0.4.
Trials: The number of trials that are run to determine the best method and combination of parameter settings.
Historical Data Smoothing Parameters: Use these parameters to get a smoother forecast from intermittent historical data.
Use Smoothed Historical Data: Controls whether the historical data is smoothed. Smoothing is typically used for weekly or finer-grained data that has many missing values. Smoothing the historical data produces a smoother baseline forecast.
Interpolate Missing Values: Specifies whether you want to smooth the data by interpolating missing values instead of by averaging. This parameter is useful when missing values indicate incomplete data instead of a lack of activity.
Median Smoothing Window: The number of time periods used in a median smoothing window to identify outliers and replace them with adjusted data values. This setting should be an odd number, so that the current time period is in the center of the window.
For monthly data, use a maximum value of 5 to prevent excessive flattening of the data. For weekly data, use a maximum of 13. Use a longer window (15 or more) for daily or hourly data.
Exponential Smoothing Parameters: Use these parameters with Single Exponential Smoothing, Double Exponential Smoothing, and Holt-Winters methods.
Alpha: A smoothing constant that determines how responsive a forecast is to sudden jumps and drops. It is the percentage weight given to the prior period, and the remainder is distributed to the other historical periods.
The lower the value of alpha, the less responsive the forecast is to sudden change. A value of 0.5 is very responsive. A value of 1.0 gives 100% of the weight to the prior period, and gives the same results as a prior period calculation. A value of 0.0 eliminates the prior period from the analysis.
Beta: A smoothing constant that determines how sensitive a forecast is to the trend. The smaller the value of beta, the less weight is given to the trend. The value of beta is usually small, because trend is a long-term effect.
Gamma: A smoothing constant that determines how sensitive a forecast is to seasonal factors. The smaller the value of gamma, the less weight is given to seasonal factors.
Trend Dampening: A constant that determines how rapidly the estimate of the trend declines over the forecast horizon. A higher value means a slower return, while a lower value means a faster return. The smaller the value, the less effect a trend has on the forecast. The value of this parameter should reflect your confidence in the trend.
Regression Parameters: Use this parameter with the linear and nonlinear regression methods.
Cyclical Decay: A constant that determines how sensitive the forecast is to deviations from baseline activity. The lower the value, the less weight is given to the deviations.
If the deviations are meaningful and ongoing for a significant number of periods, then set the value lower. If the deviations are fairly random and unconnected from adjacent time periods, then set the value higher.
Parameter Value
The current value of the setting, which you can edit.
Reset to Default
Returns all values to their original settings.
Forecast Parameter Description
Displays a description of the currently selected Forecast Parameter. To display the description of a different parameter, select that parameter in the Forecast Parameter list.
Copyright © 2003, 2007, Oracle. All rights reserved.