In order to study the robustness of the optimal orders and expected
gain, our paper considers the newsboy model and compares the optimal
orders obtained from several distributions of the demand, among them
the lognormal, normal and triangular ones. We have shown that using
these models embodies extra assumptions that are likely beyond the
actual knowledge we can extract from the data usually accessible in
inventory problems. For example using a Gaussian assumes a low 
and no skewness, while using a lognormal law assumes a non obvious
relation between and skewness.
We have also given a closed form of a bound for the cost of uncertainty,
which appears to be linear in 
and the unit selling price.
In discussions about robustness, the triangular model appears to be
useful, since it allows to reach 
and skewness 
to reach 
together with a more realistic appearance than the
two-points model. In any case, it is not clear if the usually used
is the best measure of the dispersion that can be extracted
from actual data and used in maximin discussions.
This constatation has its importance when investigating how prices are fixed by the market for risky products : in the long run, the sellers will reorient their activity if they don't obtain in the average at least the average remuneration for their capital. Thus the final price must contain not only the costs and the usual remuneration, but also (at least) an insurance for the risks[3].