Hello Prasnop,
Top-Down forecast is to use statistical forecasting on more aggregated data(like Region in the example given by DB) and then to do disaggregation(Proportion Factors) to generate the lower-level forecasts.
Alternative approach is Bottom-up forecast in which you run statistical forecasting directly to the lowest-level of detail(like Region+Customer+Material level in the example given by DB) and construct aggregate-level forecasts by summing the lower-level forecasts.
I would like to share few important points which you should consider to decide forecasting approach and level on which you should forecast.
( 1 ) . Deciding upon the lowest level at which to generate statistical forecasts(bottom-up) and deciding how to disaggregate a statistical forecast to lower levels(top-down) can have a major impact on forecast accuracy.
( 2 ). Typically(but not always), Forecast is likely to be less accurate at a lower level and it will also complicate the overall forecasting process. As you know, when we disaggregate data, the result is lower volume, less structure/pattern to the data set which will make harder to forecast using statistical models and reduced forecast accuracy (or just a flat line forecast).
( 3 ). Besides, we should always keep the CVC structure (forecasting hierarchy) as simple as possible, and make sure we do not add any unnecessary levels/characteristics in MPOS.
Hope this will help.
Thank you
Satish Waghmare