The Long-Run Australian Consumption Function Reexamined: An Empirical Exercise in Bayesian Influence

This paper reports an empirical application of new Baynesian methodology to Australian data on consumption, income, liquid assets and inflation. The methods involve the use of objective model based reference priors and objective posterior odds test criteria. The paper provides an overview of this methodology, which is based on recent work by the author (1991) and joint work with Werner Ploberger (1991) and Eric Zivot (1991). The empirical application involves tests of nonstationarity and cointegration in the data and various long-run model specifications are studied in detail. Bayesian empirical results are presented alongside well-known classical tests and are shown to provide especially useful evidence in cases where the classical test results are mixed. Our empirical results show that real private consumption expenditure and household disposal income are not cointegrated either in real or nominal terms. Instead we find strong empirical support for the inclusion of an inflation or relative capital loss measure in the Australian consumption function. Suitable measures of these variables are constructed and a final specification is recommended which yields a long-run cointegrating relation that is empirically compatible in real and nominal terms.