The following case study:
Planning and forecasting in a volatile setting by Amy Wheeler, Nina Weitkamp, Patrick Berlekamp, Johannes Brauer, Andreas Faatz and Hans-Ulrich Holst, published in Rob Hyndman's Forecasting-book (http://www.otexts.org/sites/default/files/fpp/ChulwalarCase.zip)
to my understanding compares (up to pseudonymization)
a) forecasts based on historic time-series data up to a point in time t
b) real-world planning data from a controlling department, i.e. controllers trying to (non-mathematically, e.g. by interviewing sales teams) forecast the real world from time t+1 on
c) the development of the real world from time t+1 on
Though comparison "only a)" to "only c)" is the core of any evaluation for time-series-based forecasts, the full comaprison a)-b)-c) does not appear very common in scientific articles. Reported methodology like
Wing Yee Lee, Paul Goodwin, Robert Fildes, Konstantinos Nikolopoulos, & Michael Lawrence: Providing support for the use of analogies in demand forecasting tasks, University of Bath School of Management Working Paper Series, 2005.18
Önkal, Sayım, Gönül: Scenarios as channels of forecast advice, Technological Forecasting & Social Change 80 (2013) 772–788
strive comparison b)-c), as they all involve concrete non-mathematical (i.e. not time-series) methods for forecasts along with some evaluation.
From what we saw, the closest match to an a)-b)-c) comparison are comparisons, when b) is explcitely a human-modified a) under research, e.g.
Franses and Legerstee: Do Experts’ Adjustments on Model-Based SKU-Level Forecasts Improve Forecast Quality?, Journal of Forecasting J. Forecast. 29, 331–340 (2010)
The threefold comparison however should show the potential of forecast automation in industry. Is there already any scientific literature beyond the last examples really tackling the full a)-b)-c)? Does the research operate with data and practises from real enterprises?
