PERFORMANCE OF THE STEIN-RULE ESTIMATORS WHEN THE DISTURBANCES ARE MISSPECIFIED AS SPHERICAL

抄録

The paper investigates the effects of misspecifying the disturbances in a linear regression model as spherical on the efficiency properties of the Stein-rule estimators. Asymptotic distribution of the Stein-rule estimator based on the OLS estimator is derived when the disturbances covariance matrix is nonscalar. The effects of non-spherical disturbances on the dominance conditions of the Stein-rule estimator is also observed. The risks under quadratic loss function of the Stein-rule estimators based on the OLS and the FGLS estimators are compared under a Pitman drift criterion.