Dynamic programming with value convexity (Q2665320)

In this paper, the authors investigate the abstract dynamic programming model in discrete time with a nonlinear state-action aggregator. Concepts of value-convex and value-concave aggregators are introduced for maximization and minimization cases. The authors derive optimality conditions to prove that the Bellman operator is geometrically stable. They show the validity of the Bellman principal of an optimality as far as the existence of an optimal policy and the uniqueness of a solution to the Bellman equation. The authors demonstrate how using their results one can solve dynamic programs with recursive preferences. Examples include Epstein-Zin preferences and an ambiguity aversion.