Title: A comparison of generalised maximum entropy and ordinary least square

Authors: Manije Sanei Tabass; G.R. Mohtashami Borzadaran

Addresses: Ordered and Spatial Data Center of Excellence, Department of Statistics, Faculty of Mathematical Sciences, The Ferdowsi University of Mashhad, Mashhad, Iran ' Ordered and Spatial Data Center of Excellence, Department of Statistics, Faculty of Mathematical Sciences, The Ferdowsi University of Mashhad, Mashhad, Iran

Abstract: The generalised maximum entropy (GME) estimation method is based on the classic maximum entropy approach of Jaynes (1957). It has the ability to estimate the parameters of a regression model without imposing any constraints on the probability distribution of errors and it is robust even when we have ill-posed problems. In this paper, we simulate two sets of data from regression model with different distribution for disturbance, standard normal and Cauchy distributions respectively. For this dataset, regression coefficients are obtained by GME and OLS methods and these techniques are compared with each other for some sample sizes. Moreover, we have used some prior information on parameters to obtain GME estimators. The estimation results of GME in the case of non-normal distributed are discussed here.

Keywords: regression model; generalised maximum entropy; GME; Monte Carlo experiment; ordinary least square; OLS.

DOI: 10.1504/IJIDS.2018.095495

International Journal of Information and Decision Sciences, 2018 Vol.10 No.4, pp.297 - 310

Received: 25 Feb 2017
Accepted: 16 May 2017
Published online: 08 Oct 2018 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article