1 GM–FIID ESTIMATOR AS REMEDIAL MEASURE OF HETEROSCEDASTICITY CAUSED BY OUTLYING OBSERVATIONS IN MULTIPLE LINEAR REGRESSION

Abstract

Outlying observations are usually responsible for causing heteroscedasticity in a data set. Heteroscedasticity occurs when the residual variances of a linear regression model are not constant. In the presence of heteroscedasticity the variance co-variance matrix of the ordinary least squares (OLS) becomes inconsistence which leads to inefficient estimation. The existing remedial for heteroscedasticity are not efficient to remedy the effect of heteroscedasticity cause by outlying observations. In this paper, a new remedial measure has been developed based on GM estimator and fast improvised influential distance (FIID) termed GM–FIID estimator to remedy the effect of heteroscedasticity cause by the presence of outlying observations. The results based on Monte Carlo simulation study show that the proposed GM–FIID estimator outperformed the existing methods (OLS, KNN and TSRWLS).