Optimal Investments Using Empirical Dynamic Programming with Application to Natural Resources.

A number of problems arise when dynamic programming is applied to investment problems. Attempts have been made to circumvent these problems by using continuous time models where the state process is generated by a stochastic differential equation. In this article, the authors assume that the state process is described by a discrete Markov time-series model with empirically determined transition probabilities. Standard computer algorithms can then be used to obtain optimal policies. They apply their methods to two examples in natural resources, one of which has previously been analyzed by M. J. Brennan and E. S. Schwartz (1985) using continuous time models. The authors show that the optimal policies depend critically on the specification of the state process and, to a certain extent, on the discretization level used in the Markov approximation. Copyright 1989 by the University of Chicago.