Author : Sherif Rushdy
Publisher :
ISBN 13 :
Total Pages : 0 pages
Book Rating : 4.:/5 (144 download)
Book Synopsis Maximum Likelihood Estimation of a Nonlinear System Dynamic Market Growth Model by : Sherif Rushdy
Download or read book Maximum Likelihood Estimation of a Nonlinear System Dynamic Market Growth Model written by Sherif Rushdy and published by . This book was released on 1981 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: Recent efforts to determine the proper role of formal statistical estimation in modeling "System Dynamics" models show that the parameter estimates derived from Ordinary and Generalized Least Squares (OLS and GLS) are highly sensitive to errors in data measurement, and likely to prove misleading if used as a basis for selection of parameter values or structural analysis. Using the framework developed in this previous work, - where a nonlinear feedback model generates synthetic data, which is then used to estimate model parameters and thus provide a basis for the evaluation of an estimation technique, - this thesis reviews previous results with OLS and investigates alternative estimation techniques. A review of both econometric and engineering techniques, together with some preliminary experimental results revealed that no econometric estimation technique proved capable of meeting the requirements of the estimation of parameters in a nonlinear dynamic feedback model in the presence of measurement noise. The only promising method, the Filtering Form of the Maximum Likelihood algorithm, was found in the engineering literature where it is being used in a growing number of applications. A general FORTRAN program was developed to implement this algorithm and was tried out on three small-scale linear and nonlinear models. The method was found to be capable of drastically improving upon the Least Squares estimates, if sufficient Knowledge about the noise statistics (which present identifiability problems if they are all to be estimated) was available. However, experimentation on Forrester's "Market-Growth" model, while still modest in size compared to many System Dynamics models (nine equations and fifteen parameters to estimate), revealed the many limitations (in particular in convergence and cost) of this algorithm, that preclude its use in socio-economic applications. In the light of the above results, alternative methods of model validation, and in particular a more formal use of sensitivity testing, are suggested for further research.