Comparison of some Bayesian heteroscedastic error models

Abstract

Heteroscedastic error models, with variance of the error term as proportional to the square of its expectation, have been considered. The distribution of the error term is assumed to follow either Normal, Lognormal or Gamma distribution. We estimate and compare these three different heteroscedastic error models from a Bayesian point of view using non-informative priors. A practical example which utilises expenditure data favours the Lognormal distribution to Gamma and Normal distributions.