Abstract
For a linear model with restrictions of the normed linear parameter to a generalized ellipsoid it is proven, that the Kuks-Olman estimator is minimax linear with respect to the matrik risk. From this follow optimal ridge and shrunken estimators.