Forecasting
Use
You can use a forecast procedure to predict the future development of key figure values. The default planning function type Forecasting in BI Integrated Planning offers a number of strategies and statistical methods for calculating forecasting future values on the basis of historical data.
Integration
The strategies and methods of this planning function type are based on the same statistical methods used in demand planning.
For more information about forecasting in demand planning, see http://help.sap.com → SAP Business Suite → SAP Supply Chain Management → SAP APO 3.1 → Application Help → Demand Planning → Demand Planning Process → Definition/Redefinition of Forecast Models → Creating a Forecast Profile → Univariate Forecasting
Prerequisites
● Historic data is available for the forecast calculation.
● The aggregation levels where you are creating a forecast planning function have to contain at least one time characteristic (for example, Fiscal Year/Periods). The forecast only works with one time characteristic. Only values for this characteristic can be changed. The other characteristics, especially redundant time characteristics, are not adapted by the forecast function. Time characteristics always need to assume consistent values to one another, however. Value 2005 for time characteristic Calendar Year (0CALYEAR) therefore matches the values 01.2005 to 12.2005 for characteristic Calendar Year/Month (0CALMONTH). If you want to forecast values beyond a single year, however, the calendar year must assume different values in the records to be forecast. The planning function does not do this. It is done by the derivation instead, which automatically fills the redundant time characteristics.
Note that you cannot add any redundant time characteristics such as Calendar Year/Month (InfoObject 0CALMONTH) and Calendar Year (InfoObject 0CALYEAR) or Fiscal Year/Period (InfoObject 0FISCPER) and Fiscal Year (InfoObject 0FISCYEAR) to the aggregation level,. Only add time characteristics with the finest level of granularity to the aggregation level.
In the examples specified, you should therefore use 0CALMONTH or 0FISCPER. The values for the “parent” time characteristic 0CALYEAR or 0FISCYEAR can then be derived automatically.
Features
Planning function type Forecast covers various univariate forecast procedures. In a forecast procedure, only the time series of the selected forecast key figure is taken into account. No further information is entered in the forecast calculation to interpret the development of the key figure.
Time Series Patterns
You can create forecasts for the following time series patterns:
Constant
The historic data is essentially constant and varies very little from a stable mean value. In the graphic below, this base value is represented by a red line:
Trend
The time series rises or falls continuously. In the graphic below, this trend is represented by a red line:
Seasonal
The values show periodically recurring peaks and troughs (on an annual basis). There is a stable mean value. In the graphic below, this base value is represented by a red line:
Seasonal Trend
This time series is a combination of the trend and seasonal patterns. The seasonal variation increases for an upward trend.
Intermittent
The value is zero at most points in the time series. The values that are not zero fluctuate around a mean value.
Forecast Strategies
The forecast strategy determines which forecast procedure is used. To choose a suitable forecast strategy, base your decision on the time series pattern. The various forecast procedures are based on the different forecast models (time series models). They therefore produce different results.
The following forecast strategies are available:
● Average
● Moving average
● Weighted moving average
● Linear regression
● Seasonal linear regression
● Simple exponential smoothing (constant model)
● Simple exponential smoothing with alpha optimization (constant model)
● Linear exponential smoothing (trend model)
● Seasonal exponential smoothing (seasonal model)
● Seasonal trend exponential smoothing (seasonal trend model)
● Croston model
● Automatic model selection
The automatic model selection forecast strategy allows you to let the system select the forecast model that best fits the time series of the historic data (see Automatic Model Selection).
If you already know that a particular forecast model is well suited to the time series pattern, or if you explicitly want to use a particular forecast model for other reasons, you can select a particular forecast model (see Forecast Strategies).
Additional Functions for Forecast Strategies
The forecast strategies offer the following additional functions and options:
● Logging Statistical Key Figures
● Ignoring Initial Zero Values
For exponential smoothing:
● Optimization of smoothing factors for exponential smoothing
For forecast models with trend components:
Gaps in the Forecast or in the Historic Time Frame
Gaps can occur in both the forecast and the historic time frame. Unlike forecasts in BW-BPS, these gaps are not ignored. This means that the selected times will have gaps when placed in a row. Gaps in the forecasting period are handled so that the values of this period are not changed. Gaps in the historic time frame are included in the forecast calculation with value 0.
Note that the times before the first forecast time belong in the past.
You want to generate forecast data for the months 1-3 and 5-7 (month 4 is handled separately). The system calculates forecast values for all months 1-7, but does not change month 4. In the BW-BPS forecast, however, only 6 forecast values are calculated and are assigned to the months 1-3 and 5-7 one after the other. For a linear trend in the forecast result with values 1010, 1020, etc., this implies the following difference between BI Integrated Planning and BW-BPS:
Handling of Gaps in BI Integrated Planning and BW-BPS
Month |
BI Integrated Planning |
BW-BPS |
1 |
1010 |
1010 |
2 |
1020 |
1020 |
3 |
1030 |
1030 |
5 |
1050 |
1040 |
6 |
1060 |
1050 |
7 |
1070 |
1060 |
Activities
To create a planning function of type Forecast, you have to perform the following steps:
...
...
1. Select the Time Characteristic for the Forecast.
Choose For Characteristic Usage. You choose the time characteristic that represents the time dimension of the forecast.
Note that there is a maximum valid time interval for each time characteristic. This can be set in the system. If you are using a time characteristic, the maximum valid time interval has to cover the entire planning timeframe.
On the General Settings tab page, you specify the value on the F4 Help and Hierarchies for Time Characteristicsscreen (transaction RSRHIERARCHYVIRT). Since this setting impacts on performance, you should keep the interval as small as possible.
You cannot include the selected time characteristic in the set of characteristics for conditions. For more information about using characteristics and condition characteristics, see Planning Functions.
2. Specify the forecast data.
Choose For Parameters and perform the following steps:
a. Select the Forecast Key Figures
Specify the key figures that you want to calculate the forecast for.
b. Specify the Forecast Time Frame
Specify the time frame for the forecast by restricting the time characteristic for the forecast. This is usually a time interval that represents the length of time for the required forecast.
If the time characteristic for the forecast is Fiscal Year/Periods (0FISCPER), the system proposes the higher-level characteristic Fiscal Year Variant. You only have to restrict this characteristic if you are using variables with processing type Customer and SAP Exitin the restrictions for Fiscal Year/Periods(for example, Current Periods).
The system ignores exceptional periods of the time characteristic Fiscal Year/Periods (0FISCPER) when it performs the calculations; values for periods of this type are not generated or changed.
3. Enter the historic data.
You specify the Historic Time Frame in the same way as the forecast time frame. The longer the time frame, the better the quality of the forecast results.
You use the Filter for Historic Data if your historic data differs from the forecast data for particular characteristics. You have to specify a single value for each of these characteristics.
This may be the case, for example, with the Version characteristic if the forecast data is in a plan version and the historic data is based on an actual version.
4. Select the forecast method and define more parameters.
Choose the required value for the Forecast Strategy parameter. Depending on the selected forecast strategy, the system proposes additional parameters. In certain cases, entering a value for individual parameters is mandatory.
The system proposes Automatic Model Selection as the default forecast strategy and offers the largest number of parameters after comparison with other forecast strategies. The greater the number of parameters, the more time-consuming the forecast calculation will be.
5. Save the planning function.