RELATIVE PERFORMANCE OF THE GRID, NEAREST-NEIGHBOR, AND OLS ESTIMATORS

This article provides various paradigms for the grid estimator, the mo st useful being a representation of the grid estimator as a combinatio n of the nonparametric nearest neighbor estimator and a parametric est imator. Hence, the grid estimator falls into the class of semiparametr ic estimators. The article used this representation to derive the rela tive efficiency of the nearest neighbor, grid, and OLS estimators. Und er statistically perfect conditions, the OLS estimator dominated the g rid estimator, which in turn dominated the nearest neighbor estimator. A Monte Carlo experiment verified the theoretical results. A second M onte Carlo experiment showed the fragility of the OLS superiority to m isspecification. The results cast light upon appraisal practice.