We compare the performance of tree-based models for solving stochastic programs to the performance of lattice-based approaches. We create a framework for such comparisons, argue why this framework allows us to make a definite statement about the relative per- formance and conduct various case studies. For the case studies, typical stochastic op- timization problems from the contemporary literature are chosen. Our results show that especially for high-dimensional problems with increasing dependency between the stages, lattice-based solution methods offer a higher expected out-of-sample payoff, surpassing the payoff of the tree-based methods by up to 50%. Moreover, we use statistical testing meth- ods to show that there is a significant difference in distributions of out-of-sample payoffs between those generated by a tree-based policy as compared to those generated by a lattice- based policy using Approximate Dual Dynamic Programming. We conclude that there is strong empirical evidence suggesting that the usage of lattice-based discretization together with the ADDP method provides the decision-maker with clear advantages.      «

We compare the performance of tree-based models for solving stochastic programs to the performance of lattice-based approaches. We create a framework for such comparisons, argue why this framework allows us to make a definite statement about the relative per- formance and conduct various case studies. For the case studies, typical stochastic op- timization problems from the contemporary literature are chosen. Our results show that especially for high-dimensional problems with increasing dependen...     »