A Characterization of the Cumulants as Continuous Moment-Based Statistics

Abstract

Cumulants are classical statistics associated with a random variable, defined as polynomial functions of its moments and distinguished by their additivity under convolution of distributions. Statistic is the name given to a function of a random variable, and a moment-based statistic is one that depends only on the moments $(\E[X^n])_{n\in \N}$. We prove a converse: any statistic depending continuously on finitely many moments and additive for independent sums must be a linear combination of cumulants. The proof uses an algebraic reformulation of the problem via the Hurwitz product and a linearizing change of coordinates. This result also follows from the more general theorem of Mattner \cite{mattner}, but our approach is elementary and self-contained.