Hello. I am new to the site and to this forum, and would like to solicit help with the composite forecasting feature within APO. Our Demand Planning team has been experimenting with composite models for about the past year, with limited success. I would like to get some advice and possibly share experiences with others who have had success in this arena. Specifically, we are attempting to use composite forecast profiles whereby APO selects one model profile or strategy (with 100% weighting) from among a group of models, as opposed to having APO develop a blended statistical forecast using a combination of multiple models/strategies having varying weights which total to 100%. The individual model is selected based on the "mode" identified within the composite forecast profile (smallest MAPE, smallest MAD, etc.). Our experience has been that APO has a decided "preference" toward single exponential smoothing models (although this is likely to be driven by the behavior of our historical data). However, even for data series which exhibit strong seasonal or trending patterns, we rarely see APO select these alternate models. Has anyone else observed similar results using composite modeling, and can anyone suggest any alternative approaches that we might try? I should note that there is a strong desire among our management team to stick with and enhance our composite modeling capabilities, as this feature was "sold" as an improvement in a recent system upgrade project.
Thank you in advance for any help in this regard.