Estimation of simultaneous equation models with error components structure (Q1210813)

This monograph discusses several methods of estimating a set of linear simultaneous equations in M jointly dependent variables, K predetermined variables and the additive error composed of three components: a time effect (varying with time but constant over units), a unit effect (varying over cross-sectional units but constant over time) and an overall effect (variable both in time and in cross-sectional dimensions). The method of feasible generalized least squares estimation of the reduced form parameters is discussed in some detail along with other methods such as generalized two-stage least-squares, indirect estimation of structural parameters in the just-identified case and a comparative evaluation of two-stage and three-stage least squares. A simple application to a model of residential electricity demand is also considered with three equations for price, demand and appliance stock, where the results are of uneven quality. Finally one must note that the statistical properties of the various estimators proposed are only asymptotic, no Monte Carlo studies are attempted to derive the finite sample distribution. This also leads to the difficulty that in case of the structural form the exact moments of the sample statistics are impossible to determine.