ABSTRACT
A nonzero mean implies a systematic component of the disturbances, which should really be included in the main part of the model. The implication of unequal disturbance variances is that individual observations on the dependent variable are no longer of equal reliability. It is therefore necessary to consider methods for testing the null hypothesis of no heteroscedasticity, against some alternative which does allow the disturbance variance to change between observations. It is more reasonable to suppose that successive disturbances are partly determined by previous values, but that there is also an additional random component to each disturbance. The crucial difference between a known disturbance parameter and an estimate is that a known parameter is nonrandom, whereas an estimate is a particular value taken by a random variable. The disturbances represent all influences on consumption except for disposable income and, in any given time period, it is quite possible that disturbances in two regions have at least some components in common.
