Empirical performance of the AIDS model: constant-coefficient versus time-varying-coefficient approaches

Abstract

The relative empirical performance of the AIDS model is examined when constant and time-varying coefficients are assumed. It is argued that it is the latter specification which will be the more appropriate. Estimation of the constant coefficient model, using quarterly date on UK aggregate consumption, finds no evidence of co-integrating relationships, and this taken to imply that any such relationship between the variables of the model has altered over time. Tests for coefficient stability also lend support to this view. The AIDS model is then estimated assuming random-walk coefficients and the results suggest that the data are much better suited to this form of the model. The theoretical restriction of homogeneity is also analysed in the context of each model specification.