Sensitivity analysis of multinomial decision trees (Q843393)

Summary: Decision trees are widely utilised to model decision-analysis problems. However, a thorough study of their properties from the sensitivity analysis viewpoint has not been offered yet. Our first task is to summarise recent results in the realm of constrained sensitivity analysis, so as to formalise the local sensitivity analysis of decision trees. We then derive a new result that allows to streamline the differentiation of decision trees and obtain derivatives at the same cost of Tornado diagrams. We formulate a new numerical procedure and illustrate it by means of a case study. We discuss the conditions that allow to extend the results to decision-making problems represented in the form of influence diagrams.