SEMAINE D'ÉTUDE SUR LE ROLE DE L'ANALYSE ECONOMETRIQUE ETC. 
T1 Le - 
sectors (*?) while the same endogenous variables lagged are 
likely to be safer in terms of inconsistency but are also likely 
to be more remote causes. The value of the result derived at 
the end of the last section is that it provides a case in which 
one set of instrumental variables is likely to dominate another 
set on both criteria. 
6.2. Available Instruments and Multicollinearu- 
There is obviously one set of variables which has optimal 
properties on several counts. These are the exogenous variables 
explicity included in the model. Such variables are (by as- 
sumption) uncorrelated in the probability limit with the disturb- 
ances, they also are in close causal connection to the current 
variables in any equation; indeed, they are some of those 
variables in some cases (*)). In the happy event that such 
exogenous variables are adequate in number and in the non- 
singularity of their variance-covariance matrix, and that no 
lagged endogenous variables appear, there is no need to seek 
further for instrumental variables to use. 
Unfortunately, this is unlikely to be the case in an economy 
wide econometric model. Such models tend to be almost seli- 
contained with relatively few truly exogenous variables enter- 
ing at relatively few places. This is especially the case if go- 
vernment policies obey regular rules, follow signals from the 
economy, and are therefore partly endogenous for purposes of 
estimation ‘). In estimating anv equation, all variables not 
(°**) On causation in general and in decomposable systems (or our block: 
recursive systems) in particular, see SIMON [29]. 
(°°) They may not cause all such variables even indirectly if the dynamic 
system is decomposable. Such cases are automatically treated in the rules 
given below. 
(**) This is to be sharply distinguished from the question of whether go- 
vernmentally controlled variables can be used as policy as opposed to 
estimation instruments. 
(61 Fisher - pag. 45