Parametric continuity in dynamic programming problems (Q5906649)

This paper derives conditions under which the solutions to parametric dynamic programming problems are continuous in the parameters. The results are summarized in Theorems 1 and 2, which state that parametric continuity holds if the reward (r), actions (a) and the law of motions (q) correspondences are all jointly continuous in the respective domains, or if they are continuous in the states for any given parameters. The usefulness of these results is shown in Section 7 of the paper with illustrative examples in the bandit problems, aggregate growth models, and the cost of adjustment models.