We considered the following semiparametric regres-. First, the generalized ridge estimators of both parameters and non-parameters are given without a restrained design matrix. Second, the generalized ridge estimator will be compared with the penalized least squares estimator under a mean squares error, and some conditions in which the former excels the latter are given. Finally, the validity and feasibility of the method is illustrated by a simulation example.

1 8 0-7803-4338-7/98/$10.00 0 1998 IEEE