SEMAINE D'ÉTUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIQUE ETC.

GJ

given equation, we have gone as far as we can. The avoidance
of multicollinearity is a necessary part of such a procedure
because multicollinearity is one sort of failure of the causal
criterion discussed above. To attempt to eliminate that multicollinearity
 which inevitably results from just those causal relations
 which are to be estimated, however, is self-defeating (%)

6.3. Rules for the Use of Eligible Instrumental Variables

We have several times pointed out that the causal criterion
and that of no correlation with the given disturbance may be
inconsistent and that one may only be able to satisfy one more
closely by sacrificing the other to a greater extent. In principle,
a fully satisfactory treatment of the use of instrumental variables
in economy-wide models would involve a full-scale Bayesian
analysis of the losses and gains from any particular action.
Such an analysis is clearly beyond the scope of the present
paper, although any recommended procedure clearly has some
judgment of probable losses behind it, however vague such
judgment may be.
We shall proceed by assuming that the no-correlation criterion
 has been used to secure a set of eligible instrumental
variables whose use is judged to involve only tolerable inconsistencies
 in the estimation of a given equation. Note that the
set may be different for different equations. Within that set
are current and lagged exogenous variables and lagged endo-(*!)

 I want to make it clear that I am not accusing KLoEk and MENNES
of attempting to do this. Their proposals are designed to eliminate multicollinearity
 in the first stage of the procedure where it is desirable to do so.
Their « Method 2 » [17, pp. 51-52] does eliminate collinearity in the second
 stage between the replaced endogenous variables and the included
predetermined ones, but this is not necessarily the same as the desirable and
irreducible collinearity among the variables in the equation discussed in
the text. It may well occur in practice that the KLoEK-MENNES proposals
lead to desirable results, although an approach using more structural information
 than does theirs seems preferable.

‘6] Fisher - pag. 49