ROBUST BOOTSTRAP FOR HANDLING HETEROSCEDASTICITY AND OUTLIERS IN THE PRESENCE OF HIGH LEVERAGE POINT

Abstract

It's fascinating how researchers are constantly improving regression analysis methods to deal with issues like heteroscedasticity. The robust MM estimator seems like a smart choice to enhance the wild bootstrap process for more accurate results in regression analysis. Researchers are debating the best bootstrap technique for dealing with outliers and heteroscedasticity in linear regression. There is a push for a more efficient and accurate method, considering the drawbacks of the Minimum Volume Ellipsoid approach. The proposal to replace MVE with ISE in the modified method is a promising step towards better speed, accuracy, and efficiency in robust bootstrapping. The specific objective of this paper is to modify the existing robust bootstrap technique (WBootMM-GM6-Liu). The methodology understudied the existing models and compared four existing bootstrap techniques with the modified version of the WBootMM-GM6-Liu to ascertain the impact of the modification. The numerical test results revealed that the modified version of the technique has the least standard errors, bias, and root mean square errors (RSME) and therefore outperforms the existing models taking into account the presence of heteroscedasticity, outliers, and high leverage points (HLPs). In the case of further research, this model can possibly be improved upon based on assessing fixed and random effects with other variables apart from those considered in this paper.