Statistical estimation of feedback models

Abstract

This paper considers the accuracy of conventional econometric techniques in estimating parameters in a feedback model. Evaluation of the estimation techni ques is accomplished by an experimental procedure in which a nonlinear feedback model generates synthetic data which is then used to estimate the parameters in the data-generating model. Comparison of the estimated and "true" parameter values provides a measure of the accuracy of the estimation technique.

Experiments with ordinary and generalized least- squares estimation (OLS and GLS) show that the parameter estimates derived from these methods are highly sensitive to errors in data measurement, especially when a model's feedback structure is not completely known. Further tests of OLS and GLS with different feedback models, as well as experimental evaluations of other estimation techniques, should help to determine the proper role of statistical methods in modeling feedback systems.