Modularity Classes and Boundary Effects in Multivariate Stochastic Dominance.

Hadar and Russell (1974) and Levy and Paroush (1974) presented sufficient conditions for multivariate stochastic dominance when the distributions involved are continuous with compact support. Further generalizations involved either independence assumptions (Sacarsini (1988)) or the introduction of new concepts like 'correlation increasing transformation' (Epstein and Tanny (1980), Tchen (1980), Mayer (2013)). In this paper, we present a direct proof that extends the original results to the general case where the involved distributions are only assumed to have compact support. This result has in turn proven useful for statistical tests of dominance without the assumption of absolute continuity. The first section introduces several concepts used throughout the paper. In the second section we recall the classic result as presented in Atkinson and Bourguignon (1982), with a slightly lighter proof using the general integration by parts formula for n dimensional Lebesgue-Stieljes integrals. In the third section we present our proof of the general result, using Riemman-Stieljes partial sums in a direct fashion that helps to clarify the role of modularity conditions and boundary effects in the sufficiency of the conditions. The last section discusses the relevance of the result and concludes.