Abstract
Econometric models are often made up of assumptions that never truly match
reality. One of the most challenged requirements is that the coefficients of
econometric models remain constant over time, in the sense that it is assumed
that the future will be similar to the past. If the assumption of constant coefficients
is not satisfied, any conclusions reached from normal (constant coefficient)
models will be biased.
Another, very closely related, contested assumption is that the functional form
(usually linear) of a model remains unchanged over time. The theory of linearity
has long been the centre of all econometric model-building. According to
Teräsvirta (1994), if linear estimates were not successful in practice, they would
have been forsaken long ago, and this has certainly not been the case. Quite the
opposite has been experienced: some very influential ideas based on the linear
relationships between variables, like cointegration analysis, have been
established. Nonetheless, there are definite situations in which linear models are
unable to grasp the underlying economic theory of the data accurately.
In developing economies like the South African economy, the notion of constant
coefficients and the assumption of linearity are far-fetched because these
economies are frequently characterized by changes in both the economic policy
and the economic structure. It is thus important to see these changes in
developing economies as providing valuable information for econometric
modelling. Incorporating these changes into models will provide not only better
forecasts, but also better information for policy analyses.
This study addresses the problem of non-linearity by applying smooth transition
autoregressive (STAR) specifications to an existing simultaneous
macroeconomic model of the South African economy. The results support the
view that non-linear models provide better forecasts than linear specifications of
equations.
Dr. I. Botha