Sources of heteroscedasticity:
Heteroscedasticity, or unequal variance, often occurs in cross-section data. For example, let us consider the savings behaviour of households. We know that savings is relatively higher among households with higher income. It is also likely that the variance of savings among high-income families may be larger than the variance among low-income families. Low-income families do not have much to save, so they cannot differ much in their levels of savings. High-income families, by contrast, exhibit wide ranges, from the miser who saves virtually all his income to the spendthrift who saves virtually nothing. Secondly, there is more freedom of choice for high income families while low income families have to spend a major part of their income on consumption goods. Clearly, a cross-section sample of data on savings and income behaviour to estimate an equation may not satisfy the homoscedasticity assumption. More generally, there is typically a problem of heteroscedasticity in cross-section studies in which there is a large variation in the size of the entities for which data are obtained. These entities could be households with widely different income levels, firms with widely different scales of operation, and nations with widely different levels of output.
There could be several reasons for the presence of heteroscedasticity: There may be certain outliers in the data which would increase or decrease error variance. Secondly, the problem of heteroscedasticity often arises because the scale of a variable varies enormously within the sample.