“…In this approach, a policy is computed by optimally solving a sequence of stochastic programming subproblems having a reduced time horizon. At each iteration, only the value of the first-stage variables is captured (we refer to [25], [24], [18], [42], for applications of this approach to different problems, to [9] for a classified bibliography of the literature and to [4] for the choice of the time horizon, stages, methods for generating scenario trees). Since in most of the cases the optimal policy of a multi-stage stochastic program cannot be computed, an evaluation of the performance of the rolling horizon approach with respect to the optimal policy is missing in the literature.…”