Scrap Value Functions in Dynamic Decision Problems

We introduce an accurate, easily implementable, and fast algorithm to compute optimal decisions in discrete-time long-horizon welfaremaximizing problems. The algorithm is useful when interest is only in the decisions up to period T, where T is small. It relies on a flexible parametrization of the relationship between state variables and optimal total time-discounted welfare through scrap value functions. We demonstrate that this relationship depends on the boundedness, half-boundedness, or unboundedness of the utility function, and on whether a state variable increases or decreases welfare. We propose functional forms for this relationship for large classes of utility functions and explain how to identify the parameters.