To increase the power of the Wald test.

To provide consistent variance estimation under heteroscedasticity.

To eliminate bias in the OLS estimator.

To ensure the errors follow a normal distribution.