Heteroscedasticity of unknown form: a comparison of five heteroscedasticity-consistent covariance matrix (hccm) estimators

Abstract

Regression model applications frequently involve violations of the homoscedasticity assumption and the presence of high leverage points (HLPs). The Heteroscedasticity-Consistent Covariance Matrix (HCCM) estimator's impact in the presence of heteroscedasticity of an unknown form was investigated in this study. The effectiveness of five variations of HCCM namely White’s estimator (HC0), White-Hinkley (HC1), Mackinnon White (HC2), Davison –Mackinnon (HC3), and Cribari-Neto (HC4) were accessed to identify the optimal Heteroscedasticity-Consistent Covariance Matrix (HCCM) estimator. In the study a simulated dataset was analysed using the Econometric View Software Version 12. The Breush-Pagan Godfery’s test for heteroscedasticity was applied and p-value of 0.0123 was obtained indicating presence of heteroscedasticity in the model. Applying the HCCM estimators and comparing the Heteroskedasticity-consistent standard errors estimates showed that HCO had 124.104, HC1 had 1189.222, HC2 had 1175.282, HC3 had 1106.94 and HC4 had 1140.707. These results reveal that HC3 and HC4 produced smaller errors compared to HC0, HC1 and HC2. The study hence comes to the conclusion that when doing inferential tests using OLS regression, the use of HCSE estimator increases the researcher's confidence in the accuracy and potency of those tests. This study therefore suggests that to ensure that findings are not affected by heteroscedasticity; researchers should use HCCM estimator but precisely HC3 and HC4, as the presented better results in comparison to others.