Simulations of full multivariate Tweedie with flexible dependence structure

Johann Cuenin, Bent Jørgensen, Célestin C. Kokonendji

Research output: Contribution to journalJournal articleResearchpeer-review


The paper introduces a variables-in-common method for constructing and simulating multivariate Tweedie distribution, based on linear combinations of independent univariate Tweedie variables. The method is facilitated by the convolution and scaling properties of the Tweedie distributions, using the cumulant generating function for characterization of the distribution and correlation structure. The method can handle both negative and positive correlations, and
allows simulation of Tweedie random vectors with given values of the mean vector and dispersion matrix, similar to the Gaussian case. The method allows simulation of multivariate distributions from many known, including the Gaussian, Poisson, non-central gamma, gamma and inverse Gaussian distributions.
Original languageEnglish
JournalComputational Statistics
Issue number4
Pages (from-to)1477-1492
Publication statusPublished - 2016

Cite this