Only three variables chosen by respondents are conducive to an analysis that is comparable with the REPIH; these are (a), (c) and (e) above.
First, the data provides weak evidence against Hall's REPIH, as the lagged value of available credit has proven to be a significant predictor of changes in nondurable consumption expenditure at the 0.20 level.(9) Secondly (and more encouraging), is the significance (as denoted by the t-statistic) of all of the variables chosen by the respondents as considerations in the nondurable expenditure decision.(10) This result appears to suggest that the methodology we have employed in the construction of our BCF may prove to be superior to the positivist approach currently pursued by many economists, or at the very least this occurrence indicates that economic research that employs a behavioral methodology is a justifiable exercise.
Looking at the results produced by two of the more recent examination of the REPIH, we note that MacDonald and Kearney produce an [R.sup.2] value of 0.29 when employing the change in the sum of disposable income as the additional independent variable.
Although the [R.sup.2] values obtained by Acemoglu and Scott are considerably higher than those produced by MacDonald and Kearney and the BCF, it should be noted that Acemoglu and Scott examine the REPIH on UK data in comparison to Macdonald and Kearney's and our employment of Australian data.
The selection of the variable, previous purchases by the respondents is an interesting inclusion, as it not only improves the econometric model by making it more dynamic in nature, but it also provides some support for the REPIH. However, as this was not the only variable suggested by respondents as an important consideration, the anecdotal evidence as well as the econometric evidence leads to rejection of the REPIH.
In summary we make three observations as a consequence of our investigation: first, the results obtained by the BCF are comparable with those obtained with the REPIH model of nondurable consumption expenditure, while the employment of the nonquantifiable variables and shorter time framed time series data could have only improved the predictability of our model, thus raising the low [R.sup.2] value obtained.
First, as observed by Mankiw and Shapiro (1985, 1986) some tests of the REPIH assumed income was stationary around a deterministic trend (for example, Plavin, 1981).
Campbell argued that under the null of REPIH, saving should encapsulate the superior information of the agent: saving should help to forecast income.
The principal focus of the last 15 years of US consumption research has been in finding out why the REPIH is typically rejected at the aggregate level.
But they find that anticipated relative prices and inflation matter for current consumption, which stir violates the orthogonality assumption of the REPIH. (20) The Dickey-fuller tests (Fuller, 1976; Dickey and Fuller, 1979) for the presence of a unit root, are then appropriate.