The case company combines SKUs into product groups and product groups into assortment groups. The methods based on advance demand information (Methods 1-3) can therefore be on a product group level, on an assortment level, or for all assortment groups together. We tested all three ways for all three methods, and for each method it turned out that forecasting on a product group level provided the best results (and for all SKUs together the worst). Therefore, we will only report the results on a product group level.
We remark that the ongoing policy of the company was actually to produce forecasts based on advance demand information, but on an assortment group level. Their method used a combination of the rules for dividing group demand (i.e. for calculating fn; n 2N)of Methods 1 and 2. It applied a different way of estimating group demand M. Instead of using (1), a planning committee consisting of mainly purchasers had to reach a consensus, also taking budget restrictions into account. Since the resulting forecasts have not been recorded, they cannot be compared to those of other methods in our empirical investigation. We do remark that letting budget restrictions play a role in forecasting obviously carries the risk of underestimations to stay within budget or over-estimations to avoid losing part of the budget (in future years).
Since having a top, mid and flop class is intuitively most appealing, only results for three categories are presented. We also tested the top-flop method with varying class sizes. However, again, this did not (signi?cantly) improve the performance. Therefore, we report results for equal-sized classes only. We note that contrary to ABC inventory classi?cation, where class A SKUs typically get special attention and their number therefore needs to be limited, class sizes do not affect the complexity of applying the top-flop method.