Stat Fcst

[Former Member]

Hi

I have a doubt. when we run the stat fcst at higher level (out of 10 chars, only on 2) the historical data is automatically aggregated upto this level. Thats why the fcst results run at higher level differ from the results of a run at detailed level ( all 10 chars).

My doubt is why am i seeing the difference in numbers in the totals ? It should be the same in totals if we run in higher level or lower levels. and why we generally ( or client wants ?) need to run at higher level (like product , location) ?

thx

venkat

[Former Member]

closing thread.

[srinivas_krishnamoorthy]

Many forecasting models are inherently non-linear. (except the constant model or linear regression) This means sum of forecasts of two different historical patterns may not be equal to the forecast of sum of two historical patterns. Usually most organizations try to find a middle ground in the heirarchy where the forecast model can have both mix and volume accuracy. Higher volume accuracy occurs at a higher level and higher mix accuracy at lower levels. A good practice is to run the models at lower levels to get the right mix followed by a forecast run at a higher level for right volume.

[Former Member]

not sure why you should be seeing the same totals

forecast is based on the patterns in history. This would differ at different levels and so would the forecast

for eg assume ABC and XYC are CVCs with forecasts and and at the agg C there is a forecast

period -3 -2 -1 1 2 3 4

agg C 50 50 50 50 50 50 50

ABC 10 20 30 40 50 60 70

XYC 40 30 20 10 0 0 0

(AB+XY)C 50 50 50 50 50 60 70

There would be a difference between C and the Aggregate

Its normal to have a debate and decision on what is the best method of forecasting. Top-down and Bottom-up forecasting are chosen depending on the business and who sells a better arguement

check this. http://bus.utk.edu/forecasting/articles/Top-Down%20_Versus%20_Bottom-Up%2011-02.pdf

there should be a quite a few other articles online as well as in books on forecasting