The Method of Moderation for Solving Dynamic Stochastic Optimization Problems

We show that using a remapped version of the decision rule for a standard dynamic stochastic optimization problem, where the mapping is defined by asymptotic limits of the rule as uncertainty goes to zero, produces a solution that is both more robust (in the sense that a stable solution emerges across a wide range of parameter values with no 'tweaking' of the algorithm) and more efficient (in the sense that a rule of given accuracy can be represented with many fewer computations) than standard solution methods.