The reconstruction of natural exponential families by their marginals
Two-dimensional natural exponential families of distributions with cumulant function \(k(\theta_1,\theta_2)\) are considered. It is shown that the following relations hold \[ \begin{aligned} k(\theta_1,\theta_2) &= k_1(\theta_1+\beta_1(\theta_2))+k_2(\theta_2)- k_1(\theta_1^0+\beta_1(\theta_2))\\ &= k_2(\theta_2+\beta_2(\theta_1))+k_1(\theta_1)- k_1(\theta_2^0+\beta_2(\theta_1)), \end{aligned} \] where \(k_1\) and \(k_2\) are the cumulant functions of the marginal distributions, \(\beta_1\) and \(\beta_2\) are some functions. Using marginals from the Morris class (i.e. the families in which the variance \(V\) is a quadratic function of the mean \(m\): \(V=Am^2+Bm+C\)) the author describes possible functions \(\beta_1\) and \(\beta_2\) and corresponding two-dimensional exponential families.
- Lectures on natural exponential families and their variance functions
- Bivariate natural exponential families with linear diagonal variance functions
- Sur une propriété des familles exponentielles naturelles de variance quadratique
- scientific article; zbMATH DE number 55912
- scientific article; zbMATH DE number 4182589
This page was built for publication: The reconstruction of natural exponential families by their marginals
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1600610)
