Forecasting

Hi,

Normally and for best performance we recommand to use 3 years history and 2 years future periods for forcasting.

And executioon of forecast is depending on customers requirement.You can execute weekly or monthly.

Regards,

Sunitha

Kalyan

That depends on the forecast model and parameters used. For example, models using seasonality and trend require more history than exponential smoothing models. Fundamentally, you need history far back enough so that the forecasting algorithm 'catches' the pattern in the underlying data.

This in turn also depends on the kind of industry you are in. For example, in a consumer products industry, even a year of history may be enough. On the other extreme, in a Spare Parts environment, you may need 5 years of history due to infrequent demand.

Rishi Menon

Kalyan,

Using an undefined acronym (FMCG) in an initial post invites misinterpretation. I will assume you are speaking about Fast Moving Consumer Goods.

There is no "normal" in forecasting. Every Marketplace, every company, every product line, every region - are all different.

First, you need to know for what purpose the client intends to use the forecast. Buying Raw materials? Projecting labor and capital equipment requirements? Alerting subcontractors? Financial Budgeting? Finished Goods stock plan? Each purpose suggests a different historical dataset.

Normally, you will get your answers from the client. Ask him how much of his sales history is relevant. If he doesn't know, then you may be able to do some 'best fit' regression analyses to see where relevancy falls off. In many consumer goods products, the product life is so short that history over 6 months is irrelevant. In other consumer products, the changes are so slow that you only need to know the economic business cycle.

In consumer goods, I have a tough time imagining any data over two years old having any meaning. Mostly, I would think quite a bit shorter.

As far as the future timestream and periodicity of the forecast is concerned, again, it depends. Forecasts are inherently risky. Most clients are reluctant to make a big investment, based only on the prediction of a computer program. You should forecast for the shortest period into the future that will meet the client's stated business requirements. If a forecast is only going to be used to procure Raw Materials, then try to cover the length of the longest lead times of your Raw Materials that will be affected by the forecast.

Forecast periodicity should only be small if a definite and convincing pattern is detected by analyzing the history. Forecast periods should be no shorter than the client's ability and inclination to change his production line run rates.

In short, I can't answer your questions. However, your client will give you all the data you need.

Best Regards,

DB49