Under heteroscedasticity and autocorrelation,
the OLS estimators, are still linear,
unbiased, and asymptotically (that is in large
samples) normally distributed, however,
OLS estimators no longer possess minimum
variance among all linear unbiased estimators.