Forecast Reconciliation

For doing aggregation and disaggregation at the product level, using the product hierarchy, we need to have proportions developed in your forecasting software so it could apply the logic of how to allocate the forecasts developed at a higher level to a lower level in the hierarchy. Simply, the proportion is a fraction that tells you how much of the total forecast needs to be distributed to multiple items at the level down.

Although it seems simple, there are a variety of ways in which you can build this proportioning logic. The methodologies differ in terms of how you calculate these proportions, and using what basis. There are also issues around how dynamic these proportions are and what happens if you manually adjust or change these proportions.The latter is very important from a tool perspective because you may need to manually adjust them often to achieve a desired result. The ability to adjust

forecasting proportions is as important as the ability to do forecast over-rides.

The three different methodologies to calculate proportions are as follows:

1. Equal proportions – Everything is disaggregated equally.
2. Constant Proportions – proportions stay constant through the forecast horizon.
3. Dynamic Proportions – proportions change based on trend and seasonality so they are different by the month of the year.

The last two methods can be based on either the historical data or the forecast data.

Here is an example to illustrate the different methodologies.

Let’s talk about a simple case where you have three levels in the product hierarchy: category, brand and SKU. Let us illustrate this with a simple example of five SKUs in a brand.

Equal Proportions Method

You could have equal proportions which means the forecast would be allocated 20% each, to each SKU underneath. This is a very simplistic solution, but probably not a very desirable solution since it ignores the relative importance of each SKU in that brand. The weighting may not be equal across all products. This as a method is rarely used.