HANDLING HETEROSCEDASTICITY IN LINEAR MODELS: HUBER-WHITE STANDARD ERRORS VS BOOTSTRAP CONFIDENCE INTERVALS

Abstract

Heteroscedasticity are one of the several violations of the assumptions of OLS. If no remedy applied, residuals with non-constant variance can lead to inaccurate and biased results. Academia has suggested a wide range of remedies to tackle with heteroscedastic residuals. In this study, we suggest another approach, bootstrapping of dataset to construct our confidence intervals. In order to compare the outcome, we look at Huber-White method and look at its performance against bootstrap intervals. Results indicate that bootstrap intervals perform equally well as Huber-White based confidence intervals. This indicates that bootstrap method, similar to Huber-White approach, can be a good remedy for heteroscedasticity.