MULTI-STAGE MEAN VARIANCE OPTIMIZATION AND WORST CASE ANALYSIS FOR INVESTMENTS

Berç RÜSTEM (Imperial College)

We consider the multi-stage extension of single period mean-variance optimization. In the presence of transaction costs and intertemporal constraints, the sequential application of the mean variance optimization problem leads to suboptimality. We consider the generation of scenario trees used by the underlying stochastic optimization framework as well as a worst-case optimal extension. The latter arises from the concern for ensuring the best performance in view of the worst-case, among numerous and reasonably plausible rival scenarios.