PLEM: A computer program for passive learning, stochastic control experiments

Attention is focused on the development of a new computer program that can be used to solve passive learning, stochastic control tracking problems that contain quadratic objective functions and linear models. PLEM, which stands for "passive learning stochastic control for reduced-form econometric models," is designed to perform comparisons of actual historical policies with "optimal" sequential policies. PLEM may be fairly termed "son of DUAL," DUAL being the active learning, stochastic control package developed by David Kendrick. PLEM differs from DUAL in many aspects. PLEM does not use a Monte Carlo generator, and it performs only passive learning stochastic control. These limitations are offset by certain advantages, however. There presently are 2 versions of PLEM, both written in FORTRAN. The first was developed on IBM Corp. mainframe computers and is dimensioned for up to 161 uncertain parameters. The 2nd version was intended for use on personal computers and handles up to 30 uncertain parameters.