Compacting multistage stochastic programming models through a new implicit extensive framework

Abstract

Deterministic equivalent models reformulate optimization problems from a computational perspective. Nonetheless, these models become computationally intractable quickly when the number of stages increase. In this context, a framework to reduce the size of scenario tree and multistage stochastic optimization problems is proposed. Scenario trees are generated using the Knuth transformation for a more compact representation. Moreover, the optimization model is described by using an implicit extensive form approach. The framework is tested in an asset-liability management multistage stochastic model with joint chance constraints, making it possible to acquire the optimal solution for large instances without any relaxation or decomposition mechanism.