ABSTRACT
The least squares estimators of the parameters will still be unbiased when error terms are heteroscedastic, but better estimates with smaller variance can be obtained by using a weighted regression, which gives greater weight or importance to those observations for which the error term has a lower variance. Thus an analysis of the residuals is of no help as a means of testing whether the expected value of the error term equals zero. The error term is generated partially by the effect of omitted variables, and in time series analysis in particular it is often the case that the level of these effects in one period is related to the level in the previous period. Consider as a final violation of the standard assumptions concerning the error term the case where the error term is not independent of all the explanatory variables included in the regression equation.
