PLEM: A Computer Program for Passive Learning Stochastic Control Experiments

Abstract

Abstract The benefits of using stochastic control methods to solve real world optimal policy problems in economics needs to be demonstrated. A potentially convincing demonstration for any problem is to compare the actual policy settings over time with those which would have been optimal had stochastic control methods been employed. If the optimal policy is more efficient at meeting targets (or meets them more accurately for the same cost) than the actual policy, then the methods are likely to find their way into the toolboxes of managers, planners, and other decision-makers. While the analytical properties of myriad economic control problems are becoming increasingly well-known, and while numerical methods and computational capacities have grown rapidly in recent years, there is a striking lack of software available to solve actual stochastic control problems. This paper reports on the development of a computer program to perform passive learning control on systems for which new information (about system parameters and forecasts of exogenous variables) is available to decision-makers each period. Upon execution the program computes the optimal path and presents a comparison table of the actual and optimal state and control variables over time, as well as the respective values of the objective functions.