Up to now, we have considered adaptive combinations of the LAD estimator with various other estimators (LS, trimmed LS, and M-estimator). The reason why we considered just LAD is that it is generally accepted as fairly robust against distribution outliers. We received manageable adaptive procedures that enabled us either to decide for LAD or for its counterpart, or eventually for a mixture of both procedures, which can be interpreted as an adjustment of LAD to the distribution shape.
KeywordsLinear Regression Model Regression Quantile Consistent Estimator Asymptotic Variance Influence Function
Unable to display preview. Download preview PDF.