Key points are not available for this paper at this time.
A class of heteroscedastic regression models is applied to the estimation of household expenditure functions using data from the 1966‐68 Macquarie University survey. Under the assumption that expenditure observations are normally distributed, the hypothesis that their variances are proportional to the squares of their expectations is investigated. In addition, the hypotheses tests derived by Amemiya (1973a) for discriminating between heteroscedastic regression models with gamma or lognormal distributions are applied to the same expenditure data.
Battese et al. (Sun,) studied this question.