Robust Estimation Using Modified Huber’s Functions With New Tails

Abstract

It is traditionally believed that robustness is obtained by sacrificing efficiency. Estimators with high breakdown point and high efficiency are therefore highly desirable. We investigate a new estimation procedure based on Huber’s robust approach, but with tail functions replaced by the exponential squared loss. The tuning parameters are data-dependent to achieve high efficiency even in nonnormal cases. In the regression framework, we show that our hybrid estimator is of high efficiency, reaching the highest asymptotic breakdown point of 50%. We have also established the n-consistency and asymptotic normality of our estimator under regularity conditions. Extensive numerical studies are carried out to compare the performances of our method and other existing methods in terms of the standard errors and relative efficiency, and the results reveal that the newly proposed method has smaller standard errors and higher relative efficiency than its competitors when the sample size is sufficiently large. Finally, we present three real examples for demonstration. Supplementary materials for the article are available online.