A Tutorial on Multivariate $k$-Statistics and their Computation

Abstract

This document aims to provide an accessible tutorial on the unbiased estimation of multivariate cumulants, using $k$-statistics. We offer an explicit and general formula for multivariate $k$-statistics of arbitrary order. We also prove that the $k$-statistics are unbiased, using Möbius inversion and rudimentary combinatorics. Many detailed examples are considered throughout the paper. We conclude with a discussion of $k$-statistics computation, including the challenge of time complexity, and we examine a couple of possible avenues to improve the efficiency of this computation. The purpose of this document is threefold: to provide a clear introduction to $k$-statistics without relying on specialized tools like the umbral calculus; to construct an explicit formula for $k$-statistics that might facilitate future approximations and faster algorithms; and to serve as a companion paper to our Python library PyMoments, which implements this formula.

References in corpus (2)

• A unifying framework for $k$-statistics, polykays and their multivariate generalizations
• A new method for fast computing unbiased estimators of cumulants