Comparison with Deterministic Approximations

In Chapter 3 we derived best linear decision rules in the presence of non-quadratic criteria. These decision rules have been shown to posses the “certainty equivalence property”, i.e. in case of linear decision rules it was shown to be not suboptimal to reduce the stochastic sequence of demand to a sequence of (optimal) demand forecasts. (However, to be precise, recall remark on p. 41). Restricting admissible policies to be linear one has to put up with a loss of optimality. This loss was studied in the last chapter for non-correlated and exponentially correlated demand sequences for which (over all) optimal results were presented. In this chapter we now study another suboptimal procedure. This procedure is usually met in practice. Instead of restricting the class of admissible policies to be linear and hence allowing for the separation property, one replaces from the outset the sequence of demand by its forecasts and then uses a (non policy restricted) rolling horizon optimization procedure [8b].