Finding optimal Forecasting Models and Exponential Smoothing Parameters for SAP APO-DP
Published 17 January 2009
FORECASTING SUCCESS STORY: Finding optimal Forecasting Models and Exponential Smoothing Parameters for SAP APO-DP
CPG/FMCG manufacturer Beiersdorf AG, Hamburg, Germany
Dr. Sven F. Crone and a team from the Lancaster Centre for Forecasting (including PhD students Nikos Kourentzes, Andrey Davydenko and RA Heiko Kausch) determined optimal forecasting profiles to be used in SAP APO-DP. The software does not provide a robust engine for model selection and parameterisation – in fact, some algorithms such as Seasonal-Linear-Regression seem to perform well (in sample) but give erratic forecasts for the future. The team from Lancaster developed a solution to find best forecasting settings in an external software and easily import these into SAP APO-DP. The solution involves computing and comparing a large number of exponential smoothing models with sophisticated algorithms for parameter search to determine the best possible model-parameter combination. This can then be exported and set up as an APO-DP forecasting profile, to be used in future runs of semi-automatic forecasting, while the accuracy and forecasting bias is tracked through alert-based planning.
For additional Information please contact Dr. Sven F. Crone, deputy director of the Lancaster Centre for Forecasting.
