Abstract
The practice of recursively computing unconditional variances as a means of checking for constancy of the variance is evaluated here. Some major weaknesses with this practice are highlighted. The first is the problem of diminishing weights-the fact that observations towards the end of the sample are given smaller weight than near the beginning. A simple way to alleviate this problem is proposed. The second weakness of the approach is that, in small samples, it is sensitive towards outliers that are not necessarily inconsistent with constancy of the variance. The third problem is that changes in the mean of the process can be confused with changes in the variance. Since checking recursive variances is a visual method, these potential problems are illustrated through a series of experiments.