Simulationsbasierte stochastisch-dynamische Programmierung

Decision trees, representing the backward recursive dynamic programming approach, are often not flexible enough to analyze real world decision problems in a risky environment. The crux is the modeling of risk. Stochastic simulation, in contrast, is a very powerful and flexible modeling tool for stochastic variables. However, the prevailing belief is that timeinterdependent decision problems cannot be analyzed by means of a forward moving simulation of stochastic paths. In this paper we demonstrate how to integrate the stochastic simulation procedure in a backward recursive dynamic programming algorithm. Using this combination of tools, the optimal strategy can be determined in a fast and efficient way. Our approach could be called "Bounded Recursive Stochastic Simulation" (BRSS).