Multistage stochastic programs are a viable modeling tool for sequential decisions conditional on information revealed at different points in time (stages). However, as the number of stages increases their applicability is soon halted by the curse of dimensionality. A typical, sometimes forced, alternative is to approximate stages by their expected values thus considering fewer stages in the resulting model. This paper shows how concepts in the available literature, such as the value of the stochastic solution, can be slightly extended to evaluate the benefit from solving a multistage stochastic program rather than an approximation obtained by reducing the number of stages. A numerical procedure for the calculation of this benefit, as well as bounds, are presented. The procedure is explanatorily applied to the well know investor problem.
University of Copenhagen, 2016