SEMAINE D'ÉTUDE SUR LE ROLE DE L ANALYSE ECONOMETRIQUE ETC. 3090 
genous variables which are unbiased conditional on the values 
of the predetermined variables (*°). It has also been suggested 
that the added asymptotic efficiency in the use of other estim- 
ators stemming from the employment of a priori information 
may in fact frequently be quite illusory as such information 
may be incorrect (1%). 
There is substantial merit in all of these arguments in 
various contexts. Fortunately, the issue is rather easy to 
decide in the context of estimation of the reduced form of a 
dynamic economy-wide econometric model. In the first place, 
such a model generally involves lagged endogenous variables. 
To estimate even the reduced form by ordinary least squares 
when such variables appear on the right-hand side does not 
yleld consistent estimates in the presence of serial correlation, 
substantially as seen above. Moreover, even if the assumption 
of no serial correlation is made, ordinary least squares still 
does not give an unbiased estimate of the parameters nor a 
conditionally unbiased forecast of the dependent variable. 
Nevertheless, one might plausibly be willing to accept such 
defects in ordinary least squares for the sake of greater effi- 
ciency. Such efficiency fails, asymptotically, however, if the 
overidentifying information on which structural estimation by 
other means is based is approximately correct as the issue is 
again one of good approximation rather than of correctness (7). 
Since restrictions on coefficients are more likely to be good 
approximations than are restrictions on disturbances concern 
ing which economic theory provides relatively little informa. 
tion, ordinary least squares is unlikely to be asymptotically 
efficient. 
On the other hand, such information as is available on the 
small sample properties of the limited-information estimators 
‘discussed below) does suggest that asymptotic efficiency may 
VV. 
“rT 
J 
- 
SH 
Fisher - pag. 15