306

PONTIFICIAE ACADEMIAE SCIENTIARVM SCRIPTA VARIA - 28

does not become consistent when one changes a current endogenous
 variable to a recent past value of the same variable even
if triangularity of the A matrix is achieved in this way. The
assumptions which lead to the consistency of least squares require
 more than this and all the same difficulties will still be
encountered even if they go unrecognized.

2.4. The Proximity Theorem and Near-Consistency

As already remarked, however, the issue of the use of
ordinary least squares (or indeed of any particular estimator)
is not whether the assumptions thereof are precisely satisfied.
Rather the crucial question is that of how closely they are
satisfied, of how those assumptions stand up as approximations
rather than as exact statements. The problem is not a discrete
one; rather it is continuous. Moreover, the question of goodness
 of approximation is itself dependent on the sensitivity of
the properties of the estimator to variation in the assumptions
thereof. In general, the less sensitive is an estimator, the greater
the tolerable deviation from the strict conditions under which it
has desirable properties.
In the present instance, our discussion has largely run in
terms of consistency. Consistency, however, is a rather weak,
although desirable property. Since ordinary least squares has
several other attractive features, we might plausibly be willing
to tolerate small inconsistencies to gain, for example, computational
 ease, small variance around probability limits, and so
forth. Tt is thus not sufficient to ask whether the assumptions
under which ordinary least squares is consistent are satisfied;
we must ask whether the fact that they are not generally satisfied
 in economy-wide econometric models is likely to be of
much Importance.
This question is formally answered by the Proximity Theo-‘61

 Fisher - pag. 12