This paper analyzes the finite-sample performance of the consistent covariance matrix estimator proposed by Halbert White under both homoskedasticity and heteroskedasticity using Monte Carlo simulation methods. It showns that this estimator can be quite biased in samples of small to moderate sizes, thus leading to associated quasi-t tests with large size distortions. The finite-sample performance of bootstrap and bias-corrected estimators is also investigated. The numerical results favor the analitically corrected estimators over the one obtained from a weighted bootstrapping scheme. The paper analyzes the finite-sample of three alternative estimators, which are defined as small variations of the White estimator. Finally, the results also show that the existence of points of high leverage in the regression matrix has a substantial impact on the finite-sample performance of the different covariance matrix estimators.
