A simulation study of least squares and ridge estimators for small samples

Abstract

In this paper we consider the Least Squares (LS) estimator (predictor) and various ridge estimators
(predictors) and report on a Monte Carlo study their small sample properties. The Monte Carlo experiment
is applied to a residential electricity demand function with data from the Greek economy. On the
basis of 2,500 replications of sample size 24 for normal disturbances we find that for the measures of
dispersion the HKB estimator appears to be superior to the rest of the examined estimators. On the other
hand the choice of alternative predictors for several measures of bias and dispersion is not clear. Furthermore,
it should be noted that the small sample properties of the ridge estimators turn out to be different
from the small sample properties of their respective predictors.