42
306) provides three motivations for determining the causal nature of the
relations of an economic system:
(1) Policy implications of any relation depend critically upon
whether the relation holds in one or more directions,
(2) Methods which are not designed to recognize the directional
nature of relations will often lead to acceptance of a rela
tion as non-directional when on the basis of available data,
only a more restricted causal relation is justified, and
(3) If we do not use techniques adapted to finding causal, as
contrasted to non-directional, relations, we may fail to find
relations which actually exist and which could be found on the
basis of available data.
If there exists a strong causal structure that is not embodied in the
structural specification of an explanatory model, the possibility of
biased and inconsistent parameter estimates exists. Bishop (1979, page
2) states that "given the potentially serious problem with simultaneous
equations bias when a simultaneous system is estimated by a single
equation method, it is important to ascertain the causal structure."
This is no less true when modelling in a dynamic lead/lag framework.
Sims (1972, page 540) notes that "most efficient estimation techniques
for distributive lags are invalid unless causality is unidirectional" in
the Granger sense. Thus, testing the implicit causal assumptions on
which most single equations or systems regressions are based is of vital
importance. Strotz and Wold (1960) emphasize that this is particularly
true when dealing with explanatory rather than descriptive "curve fit
ting" models.
The direction of causality as dictated by the theory is a debate-
able topic. Colclough and Lange (1982) express a theoretical basis for
questioning the direction of causality. They state that