SEMAINE D'ETUDE SUR LE ROLE DE L’ANALYSE ECONOMETRIOQUE ETC. 487

The median turns out to be of the order of 0.9, which
means that in about 50% of all cases the value of the objective
function (6.1) was reduced by the strategy approach by at least
10% (when our yardstick is the distance from naivety to perfection).
 Whether this is to be considered as much or as little
is a question of taste; the present author must confess that
he is not impressed by this figure. But a more important
question is whether the strategy approach leads to sufficiently
superior outcomes in general, and this can only be taught by
future research. The results of Table 4 suggest further that the
dispersion of the individual relative performances tends to decrease
 when the standard deviation of the €’s becomes smaller.
This is a somewhat surprising result if it would be true; to find
out whether it is true or not requires a larger experiment.
Simulation experiments are useful to show the random variability
 of outcomes whenever the model on which they are
based is too complicated to be handled analytically by the
ordinary methods of probability theory. The model used here
cannot claim to be complicated, but, in a more general con-‘ext,
 it is not difficult to think of other models in development
programming which do have this feature. Even in this simple
case, however, the method illustrates clearly something which
is obvious to those who think in probability terms but which is
not so obvious to those who think in terms of nonstochastic economics;
 viz., that a procedure which is good in general is not
necessarily good (at least not good in a limited time period)
in all individual cases. For example, the relative performance
measure is above I in almost 20%, of all cases. Thus, if we

mation can be improved upon if we replace b by the reciprocal of 1 minus
che x,_, of the fixed-plan policy, because it turns out (in the present
application at least) that the fixed-plan savings ratios do not differ too
much from the corresponding savings ratios of most of the strategy cases.
This alteration would imply that the multiplicative coefficients of eq. (2.9)
“ease to be constant over time, which makes the computations more comp
cated, but probablv not in a hopeless manner

Theil - pag. 2-