A Comparison of Mixed and Ridge Estimators of Linear Models

Hüseyin Güler, Selahattin Kaçıranlar

Communications in Statistics - Simulation and Computation · 2008 · 18 citations · 28 references

Concepts

Abstract

Abstract The presence of autocorrelation in errors and multicollinearity among the regressors has undesirable effects on the least squares regression. There are a wide range of methods, such as the mixed estimator or the ridge estimator, for estimating regression equations, which are aimed to overcome the usefulness of the ordinary least squares estimator or the generalized least squares estimator. The purpose of this article is to examine multicollinearity and autocorrelation problems simultaneously and, to compare the mixed estimator to the ridge regression estimator (RRE) by the dispersion and mse matrix criterions in the linear regression model with correlated or heteroscedastic errors. Keywords: AutocorrelationMean square errorMulticollinearityRidge regressionMixed estimationMathematics Subject Classification: 62J0562J07 Notes 1Actually this matrix is but for notational simplicity, we call it V. 2In some cases, mse of mixed estimator is bigger than GLSE's. However, this issue is due to Monte Carlo sampling error and it is theoretically impossible to obtain this. Therefore, these results should be interpreted as mixed and GLS estimators have equal mse for these cases.

References

28