Bottom-up forecasting involves predicting members at the lowest levels of the dimension hierarchies and optionally rolling up the results to higher level summary members.
Top-down forecasting involves predicting members at the summary levels of the dimension hierarchies and optionally spreading the results down to lower level members. This type of forecasting is useful when historical data is not available for lower level members, or when top level predictions are being used to “drive” the results down to lower members.
Note: | Forecasting results between bottom-up and top-down methods should be close, but predictions on lower level members are the most accurate since the individual trends and patterns of the data are preserved in the prediction process. |