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

+51

unlikely that economy-wide models have entirely triangular A matrices,
 nobody can disagree, and what he says is even an understatement.
 On the other hand, if matrix A is made nontriangular the
system becomes interdependent, and then we are confronted with
the snag of a possible discrepancy between the two notions of expected
 values for the current endogenous variables. In other words,
one kind of approximation has been replaced by another, and the
arguments at issue do not indicate which approximation is prefer
able. This question brings me to the third point.

(ii) Prof. FISHER’s considerations for or against CC- and IDsystems
 are on the whole of a theoretical nature. For my part 1
would instead make a plea for comparisons of an empirical nature.
Above all, my plea is for studies where one and the same observational
 material is used for the testing of fullfledged CC- and IDmodels.
 Such comparative studies are not as yet in the picture, so it
is as yet an open question which type of model performs best in
actual applications. In this connection there is the important question
 what criteria should be used when comparing and evaluating
the performances. Predictive tests here come to the fore as the real
touchstone for the validity of a model; that is, tests where the model
as constructed on the basis of past experience is exploited to make
forecasts about the unobserved future, and where in due course the
forecasts are confronted with the actual developments

FISHER

Professor WoLD has long been interested in the question of what
are desirable properties of econometric models. This is an interesting
question and his work and his present comments cast considerable
light on it. As an economist, however, I think one ought to be
interested not so much in the question: « What are good properties
of econometric models? » but rather in the question: « What are
the properties which good econometric models have? » In other
words, one does not design a model so that it will have desirable

[6] Fisher - pag. 67