International Journal of Forecasting 25 (2009) 32–34 www.elsevier.com/locate/ijforecast Discussion Reply to Commentaries by Flores, ¨ Onkal and Sanders Robert Fildes a,∗ , Paul Goodwin b , Michael Lawrence c , Konstantinos Nikolopoulos d a Lancaster University Center for Forecasting, Lancaster University Management School, Lancaster LA1 4YX, UK b School of Management, University of Bath, BA2 7AY, UK c Australian School of Business, University of New South Wales, Sydney, Australia d Manchester Business School, Manchester M15 6PB, UK We thank the three commentators for their observa- tions on our paper, particularly their recognition of the importance of the problem. Judgmental adjustments are common, much more common in fact than the heavily researched area of forecast combining (Fildes & Goodwin, 2007), and it is important to note that they have a different structure, in that the adjustments are conditional on the outcome of the current statisti- cal forecast. Thus, we believe the area provides a vital area of study, both for its practical implications and for its research challenges. In this reply we discuss some of the issues that the commentators have raised. Flores seeks clarification of our role in the forecast- ing meetings that we attended in the participating com- panies. Since our objective was to carry out research into the forecasting processes used by these compa- nies, we attended strictly as observers and played no active part in the meetings, although in negotiating access we had agreed to provide some recommenda- tions at the project’s end. Only then did we feed back our findings to the participating companies and pro- vide advice on how they could improve their processes DOI of original article: 10.1016/j.ijforecast.2008.09.001. ∗ Corresponding author. E-mail address: r.fildes@lancaster.ac.uk (R. Fildes). based on both the overall results of the project and the company-specific analyses. Flores also asks how we selected the key SKUs. We, in fact, used the entire data sets that the companies supplied us with, eliminating (as we point out) only those series where forecasts, or actual demand levels, fell below 10 units. The percent- age of series eliminated was 3.7%. We have since anal- ysed forecasts for the products that had intermittent demand in one of our companies in a separate paper (Syntetos, Nikolopoulos, Boylan, & Fildes, in press). Although the results for these products turned out to be broadly similar to those reported for the higher volume products, as Sanders points out, small judgmental ad- justments to statistical forecasts of zero demand were generally beneficial. We therefore agree with Sanders that intermittent items seem to warrant separate treat- ment in studies of the effectiveness of judgmental in- tervention. As to Flores’ question about how the SKUs were selected for adjustment, we observed two major triggers: first, a large recent error (sometimes with an ‘out-of-control’ signal in the FSS) and second, when the forecaster had received relevant additional information such as a forthcoming promotion. The empirical evidence from this study suggests that this 0169-2070/$ - see front matter c  2008 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.ijforecast.2008.09.006