GENERAL MOMENT AND QUASI-MAXIMUM LIKELIHOOD ESTIMATION OF A SPATIALLY AUTOCORRELATED SYSTEM OF EQUATIONS: AN EMPIRICAL EXAMPLE USING ON-FARM PRECISION AGRICULTURE DATA

This paper compares two estimation methods of a three-stage least squares (3SLS) system of equations, corrected for spatial autocorrelation. The modeling approach is novel in that it is an extension of Anselin's (1988) seemingly unrelated regression (SUR) space-time spatial error model for panel data. An empirical comparison of the quasi-maximum likelihood (QML) estimation of the equation system, and Kelejian and Prucha's general moments (GM) estimation approach is presented. The model and estimation procedures introduced in this study are easily extended to other economic, agronomic, or biological models that must incorporate spatial and temporal effects in the model specification, and overcome simultaneous equation bias. The empirical example used in this study falls in the realm of production economics: on-farm production data is used to optimize input rates across time and space. This data is the product of on-farm, site-specific manure management research at the University of Minnesota.