Algorithms for solving SURE models
by
Erricos J. Kontoghiorghes
University of Neuchatel, Switzerland
Coauthors: Paolo Foschi (University of Neuchatel, Switzerland)

The basic computational formulae for deriving the estimators of Seemingly Unrelated Regression Equation (SURE) models involve Kronecker products and direct sum of matrices that make the solution of the models computationally expensive even for modest sized models. Therefore the derivation of numerically stable and computationally efficient methods is of great importance.

An iterative procedure is used to obtain the feasible estimator of the coefficients when the disturbances' variance-covariance matrix is unknown. Initially, the regression equations of the SURE model are assumed to be unrelated, that is the correlation among contemporaneous disturbances of the model is ignored, and then, the disturbances' variance-covariance matrix is estimated based on the residuals of the estimators. At each iteration the estimator of the SURE model comes from the solution of a Generalized Linear Least Squares Problem.

Computationally efficient algorithms are proposed to derive the iterative feasible generalized least squares estimator of SURE models when the regression equations have common exogenous variables. The algorithm has a basic tool the generalized QR decomposition. Parallel strategies which exploit the special structure of the exogenous matrices are also considered.

Date received: January 28, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caeb-31.