Ridge estimation in linear models with heteroskedastic errors

This paper presents a number of heteroskedasticity ridge consistent covariance matrix (HRCCM) estimators in order to develop a new version of mean square error criterion (denoted by HMSE) for comparing biased estimators in the presence of both collinearity and heteroskedasticity of unknown form. New methods to choose k (designated by k*) are also proposed and examined via Monte Carlo simulations (1000 replications). The Monte Carlo results reveal superiority of the estimator aGRR (k* mgh) over some well-known biased estimators by means of trace (HMSE) criterion when values of a number of factors that may affect their properties have been varied.