Maximum Entropy Optimization
by Anthony W. Warren, Ph.D. and Jack K. Hutson
The Maximum Entropy Method (MEM) is a promising new system for short-term Spectrum Analysis
and trend forecasting. In this article we discuss the process of obtaining an optimal set of MEM
coefficients using the author's code as adapted to the Compu-Trac system.
The basic theory and terminology associated with the MEM method are discussed in a Technical Analysis
article "An Introduction to Maximum Entropy Method (MEM) Technical Analysis", published
January/February 1984. To review some of the major points, MEM is a system which can provide high
resolution Spectra for identifying the dominant data cycles, using relatively short time series. (Fourier
Analysis, in contrast, gives best results when applied to time series on the order of six-months to a year or
longer.) Moreover, the MEM fit coefficients may be used to produce a data forecast which is consistent
with the identified cycles. However, this system is more complex to use than Fourier Analysis, since its
application requires following an optimization process. In the author's system, one can select either
automatic optimization or manual fit selection for optimizing the MEM data fit.