The Regrettable Loss of Mathematical Molding in Econometrics - Mechanism and Causality in Biology and Economics - page 76

the 1950 monograph of the Cowles Commission, one finds only one reference to
“autonomous relation,” namely, to Haavelmo's chapter “Remarks on Frisch's
Confluence Analysis and Its Use in Econometrics.” In it he described an autono-
mous relation as a relation that “would hold regardless of whether or not other
economic relations were fulfilled” (Haavelmo 1950 , p. 263). This is again the case
in the follow-up of the 1950 monograph, namely, the 1953 monograph Studies in
Econometric Method . 7 There is only one reference of “autonomous equations” in
the index, in Girshick and Haavelmo's chapter “Statistical Analysis of the Demand
for Food: Examples of Simultaneous Estimation of Structural Equations.” But now
“structural” and “autonomous” seem to have converged. 8
Why is it that we are interested in one particular member of this infinite set of true systems?
It is because, in setting up the original model, we believe that there is one particular system
of equations that is a system of autonomous ,or structural equations, that is, equations such
that it is possible that the parameters in any one of the equations could in fact change, e.g.,
by the introduction of some new economic policy, without any change taking place in any
of the parameters of the other equations. (Girshick and Haavelmo 1953 , p. 106)
The reason why researchers at the Cowles Commission believed that the struc-
tural equations were autonomous is that the empirically found relationships may be
simpler than theory would suggest. This could lead researchers to overlook poten-
tial influences. Moreover, there may be factors that were not only overlooked
because they were not revealed empirically but were also not yet accounted for in
theory. However, as passive observers “we cannot clear the data of such 'other
influences', we have to try to introduce these influences in the theory, in order to
bring about more agreement between theory and facts” (Haavelmo 1944 , p. 19).
Thus, it was assumed that the problem of autonomy could be avoided by building
models to be as comprehensive as possible.
5 Conclusions
Originally, in business-cycle analysis, whether a potential causal factor was added
to the business-cycle mechanism depended on whether it was theoretically as well
as statistically and mathematically significant. Mathematical significance of a
causal factor depended on considerations of whether the model containing that
factor generated the appropriate facts about the phenomenon. Because feedback
from the phenomena was cut off, the Cowles Commission approach was not to
discover or test, but only to identify and measure.
7 Both monographs are considered as containing the main body of the Cowles Commission's
theoretical results (see Christ 1994 , p. 32).
8 For a more sophisticated account of this convergence, see Chao ( 2009 ), where he distinguishes
between the invariance view and the theory view. Autonomy is equivalent to invariance but “does
not say anything about the constitution of a system” (p. 71). So, the convergence is best described
as the “invariance view of structure.”