SEMAINE D'ÉTUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIQUE ETC. 
380 
of a structural equation, although not when used to estimate 
the equations of the reduced form. 
This argument is not sufficient, however, to dismiss ordinary 
least squares from consideration as an appropriate estimator 
in large econometric models. In the first place, there is the 
question emphasized by H. WoLb (°) as to whether such models 
really should be simultaneous given the nature of causation. 
Second, the issue is not of the yes-or-no variety as it is often 
made to appear; rather, if the model is such that correlation 
between the disturbance term and the explanatory variables 
in the given equation can be appropriately assumed to be small 
(rather than zero) or if the variance of the disturbance is known 
to be small, then least squares will be almost consistent (°). 
One may then be willing to accept the small inconsistencies 
involved for the sake of the other properties of the estimator, 
principally its relatively small variance around its probability 
limit. We must therefore go on to ask when this is likely to 
happen and when the assumptions of WoLD’s recursive model 
are likely to be approximately satisfied. 
2.2. Recursive Systems and Necessary Assumptions 
Suppose. that the model to be estimated is: 
(2.1) 
Ay,+ By, ,+Cz,+u, 
where u, is an m-component column vector of disturbances; 
y, is an m-component column vector of current endogenous 
variables; z, is an #-component column vector of exogenous 
variables (known at least to be uncorrelated in the probability 
limit with all current and past disturbances); A, B, and C are 
constant matrices to be estimated; and (I - A) is nonsingular, 
{ WoLD and JURÉEN [34, 50-51] and other writings 
(*) Worn and FaxÈr lag" 
wo 
Fisher - pag.