Error variances of OLS is assumed constant
(homoscedasticity). However, when
error variances are not constant, there is
heteroscedasticity. When there is heterscedasticity,
OLS estimations places more
weight on the observations with large error
variances than those with smaller error variances.
The OLS parameter estimators therefore
are unbiased and consistent, but they
are not efficient, therefore the consequence
is similar to those of autocorrelation.