On Wed, 2 Mar 2022, Marcin Błażejowski wrote: Well, they are, up to a point. Let me expand your example: <hansl> nulldata 10 eval 1/0 eval {1/0} eval {-1/0} scalar a = 1/0 matrix b = { 1/0 } series c = 1/0 series d = a series e = b series f = mshape(a, $nobs, 1) series g = mshape(b, $nobs, 1) print a b print c .. g --byobs </hansl> I would say that what we do is try to reach some compromise between IEEE 754 (see https://en.wikipedia.org/wiki/IEEE_754-1985) and what an applied economist would consider "natural" in the context of data analysis. So, for example, we take the liberty to map ±inf to NA when this is appropriate. But as always, Allin is more authoritative than me, so let's see what he has to say. ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------