Authors: Graeme J. Doole
Addresses: Waikato Management School, Department of Economics, University of Waikato, Private Bag 3105, Hamilton, New Zealand
Abstract: The generalised maximum entropy (GME) estimator provides a flexible means of information recovery from ill-posed estimation problems. However, coefficient estimates are sensitive to the exogenous support bounds defined for coefficient and error terms. This paper describes a new estimator that identifies informative support bounds, prior to the implementation of GME regression. These bounds are estimated using interval-valued mathematical programming in a way that is data-based, replicable, and robust. The superiority of the new estimator over various alternatives is demonstrated with a series of non-trivial Monte Carlo simulations involving different degrees of multicollinearity, sample sizes, and error structures.
Keywords: maximum entropy; support bounds; ill-posed problems; multicollinearity; low sample size; interval-valued optimisation.
International Journal of Computational Economics and Econometrics, 2018 Vol.8 No.2, pp.129 - 143
Accepted: 22 Dec 2016
Published online: 09 Apr 2018 *