Composite Forecast Method
A forecast method that generates a blended forecast by combining (either by averaging or weighted averaging) multiple forecasts from different forecast streams.
In most cases, composite forecasts are generated by blending two or more statistical and causal forecasts. This generates better forecasts, which means forecasts that are more accurate. These forecasts can include the following:
• Statistical Time Series Forecasts
• Causal Forecasts
• Leading Indicator
• Scheduled Event
The formula for generating a composite forecast is as follows:
weightfactor1 * Fcst1 + weightfactor2 * Fcst2 + weightfactor3 * Fcst3 …
=Σ(weightfactor(i) x (forecastamount(i) + forecastschamount(i) + forecastnramount(i)))
Where:
• weightfactor(i) = Weight factor assigned to each forecast stream - can vary by time
• Fcst(i) = Total forecast in IPCS_FORECAST_DETAIL, including adjustments
• i = number of forecasts streams with UsedinComposite = y
Values for these fields are calculated when the Composite forecast method is used:
• MAD
• MAPE
• RMSE
• Tracking Signal
 |
Use these recommendations for implementing the Composite forecast method
• The Forecast Netting process rolls up forecasts from composite forecast streams to forecasted_data if UseInTotal flag is set.
• HistoryAvg , HistorySD, FoercastErrorSD, and ForecastSD are computed for forecast_detail and forecasted_data.
• Users cannot make forecast adjustments to composite forecast streams. Mass adjustments to composite forecasts on the Forecast Analysis page is not supported either.
|