SEMAINE D'ÉTUDE SUR LE ROLE DE L ANALYSE ECONOMETRIQUE ETC. 
455 
fact been adopted in economy-wide models in the past. My proposa. 
is to take each equation (and indeed each endogenous variable 
within each equation) and to get a group of instrumental variables 
which are particularly appropriate for it. One does this by consid 
ering the causal structure of the system. The block-recursive system 
— in which the full system is divided into sectors and subsystems 
— is used in my paper to point out that it is safer to use some 
endogenous variables in some places than in others so that the 
eligible list of instruments is different for different equations. 
HAAVELMO 
I wonder whether FISHER would agree to a rough summing u 
of the situation that might help people who are not so much inte- 
rested in technical details and who wonder what we get out of all 
this. The summing up could be something like this: We have 
small samples in practice, so that large sample properties of esti: 
mates are probably not too interesting. Then, if we had a clear 
situation as far as identification is concerned, that is to say, if we 
need no subtle tricks in order to ensure identification; and secondly, 
if by means of mere inspection we can find that there are exogenous 
variables which vary a great deal as compared with the disturbances 
u, and the exogenous variables are really uncorrelated with the u’s 
in various directions, both simultaneously and recurrently; then we 
get reasonably good estimates even without refined methods. Now 
it is in the in-between cases that we may perhaps gain the most by 
being very particular about the estimation methods chosen. This 
would intuitively be the conclusion that one who works in practice 
with these things would draw. This does not of course in any 
way reduce the importance of stringent work on estimation methods, 
but I think it might — if it is reasonably correct interest those 
who want to know what the practical conclusion .- 
69 
ag. 
isher - p 
vo] F