Propriétés en échantillon fini des tests robustes à l'hétéroscédasticité de forme inconnue

In this paper, I investigate the finite sample performance of a test robust to heteroskedasticity of unknown form, based on the consistent covariance matrix estimator proposed in Eicker (1963) and White (1980). The simulation results suggest that, as often used in practice, this test could be unreliable and inefficient, even if the sample size is large. They suggest that reliable and more efficient inference can be obtained if a heteroskedasticity-robust test is computed with the restricted residuals and an appropriate bootstrap method.