Multivariate Bernoulli Distributions with Actuarial Applications

Moderator: Tom Salisbury, Professor Emeritus, York University

Multivariate Bernoulli distributions are essential in the modeling of binary data in a wide variety of contexts, such as actuarial science, quantitative risk management, machine learning, natural language processing, and bioinformatics. In this talk, I will present recent results on two topics related to multivariate Bernoulli distributions.

 

In the first part of my talk, I will consider tree-structured Ising models, a class of undirected graphical models for Bernoulli random vectors. We introduce a stochastic representation of the components of the Bernoulli random vectors such that the marginal distributions remain fixed. That representation has many advantages: to find the Pearson's correlation coefficient for any pairs of components of Bernoulli random vectors; to design an efficient sampling algorithm; and to investigate properties of the tree-structured multivariate Bernoulli distributions. We also derive an analytic expression for the joint probability generating function of the Bernoulli random vector. The latter is used to build efficient computation methods for the sum of the components of the Bernoulli random vector.

 

In the second part, I will discuss negative dependence for Bernoulli random vectors. We will provide the essential tools of negative dependence and extremal negative dependence in a common language and the framework of multivariate Bernoulli distributions. We will characterize sigma-countermonotonicity and study the strongly Rayleigh property within this class.  We will illustrate those notions using the class of conditional Bernoulli distributions.

 

To conclude, I will present examples with actuarial applications of the results from the two topics.