For any optimization problem, the quality of the solution depends heavily on the quality of the input data. In stochastic programming, the input for the models is usually given in the form of scenario trees based on an underlying statistical distribution. We look at examples where the discretization of the underlying distributions gives unstable results that are too optimistic and turn out to be infeasible when tested using the underlying distribution. Examples and suggestions for dealing with such problems are presented.

Keywords: scenario generation, stability, stochastic programming, logistics