Show TOC

Function documentationMLR Forecast with POS Data

 

Nowadays, point of sale data (POS data) can be processed in such a way that it can be used in Demand Planning (DP) to create forecasts. This allows a supplier to create a short-term statistical forecast which, in contrast to other statistical forecasts, achieves more exact results, since it also considers, in addition to the resulting deliveries and the POS data, the customer’s inventory fluctuation.

Prerequisites

The prerequisites for the forecasting method are as follows:

  • The data for the deliveries in the past and the POS data must be available in weeks.

  • The forecast horizon may not exceed 12 weeks.

  • For the constant model, you require a time series of at least 25 weeks in the past and, for the seasonal model, you require a time series of at least two complete seasonal cycles. The two seasonal cycles together must cover at least 25 weeks.

Features

This forecasting method executes multiple regression analysis (MLR) several times and additionally uses either the constant model with first-order exponential smoothing or the seasonal model. The seasonal model used here is slightly different from Winter’s seasonal model. With this method, you can create a forecast for up to 12 weeks in advance.

The forecasting method first calculates the weekly changes in the resulting deliveries and in the POS data. The system uses multiple regression analysis to then determine the quantity-based relationship between the data, in order to determine the inventory fluctuation in the past. Based on this information, the system creates a statistical forecast for the whole forecast period, depending on the statistical model that you have selected.

An additional regression analysis aims to quantify the relationship between the deliveries in the past, the statistical ex-post data, and the inventory fluctuation in the past. The final forecast result is a statistical forecast that considers future inventory fluctuation.

In the first run, the system calculates forecast data for the first week. Depending on the forecast horizon, the system continues calculating until it has determined the forecast data for the first four weeks. If the forecast horizon extends past four weeks, the system uses the data from deliveries in the past along with the statistical forecast data that has already been calculated and the POS data together with the average values from the first four weeks as the basis for the forecast calculation as of the fifth week.

For the MLR forecast with POS data, you can also use the mean absolute percent error (MAPE) error measure. In this case, the system calculates the error measure once for the whole forecast method and once for the statistical forecast itself. If the error measure for the forecast method is greater than the one for the statistical forecast, the system generates a message, for example, during background processing in the job log. The system can also issue the message as an alert.

The MLR forecast with POS data supports lifecycle planning. For the forecast with like profiles, the system uses the corresponding historical data and the POS data of the reference objects.

Activities

To make settings for this forecast method in the MLR profile, on the SAP Easy Access screen, choose Start of the navigation path  Advanced Planning and Optimization Next navigation step Demand Planning Next navigation step Environment Next navigation step Maintain Forecast Profiles End of the navigation path. Create a new master forecast profile or use an existing profile. You make settings for the MLR forecast with POS data on the MLR Profile tab. In the Profile type screen area, choose the corresponding MLR method.

For more information about the settings for this forecasting method, see MLR Profile

.