SEMAINE D ETUDE SUR LE ROLE DE L ANALYSE ECONOMETRIOUE ETC. 
1216 
macroeconomic level, The purpose of the device is much more prag- 
matic, in the direction of clarifying the problem, articulating different 
political views by specifying alternative targets and alternative instru- 
ments, targets and instruments that can be modified from time to 
:ime, the limited aim not being solutions of absolute optimality, but 
rather solutions that make workable compromises. If I understand 
correctly, Professors DorrFMAN and FriscH both adopt this prag 
matic view. Thus it is seen that the pragmatic approach is far from 
uniform, and that it leaves room for widely divergent views about 
fundamental problems. 
Coming to my second point, I am not altogether happy about 
‘he points in Professor FRISCH’s papers and his presentation where 
ne limits himself to deterministic approaches. It is always necessary 
to simplify in model building, and it is also granted that deterministic 
assumptions are sometimes adequate; however, I feel uncomfortable 
when Professor FRISCH says that deterministic assumptions are a dash 
forward — I think it is a dash backward. It is an illusion if you 
selieve that you can get rid of difficulties in economy-wide models by 
supposing that you are so precise that there are no unexplained 
residuals in your approach. To reduce the residuals you must stratify 
in great detail so as to obtain homogeneous cells in your statistical 
tables; when the cells become small, however, the law of large num- 
bers ceases to work and instead of more deterministic regularity you 
will run into more randomness and irregularity. The problem of the 
model builder is to strike a sound balance between the gain in infor- 
mation given by a finer stratification, and the loss in information 
when the law of large numbers is weakened. In a sense the residuals 
are the back side of the medal of large numbers; they will however 
do no harm if they are treated as stochastic variables, and if the 
model takes them into account in such manner that the operative use 
of the model is in accordance with the mathematical rules for operat- 
ing with random variables. This last principle takes care of the 
pitfalls when a deterministic model is stochasticized. Now as far as I 
can see the deterministic models considered by Professor FRISCH are 
not in the danger zone in this respect. At least for ordinary input 
“17] Frisch - pag. 23