Development of Alternative Factor Copula Models

Develop alternative factor copula models for the multivariate zero-inflated mixed Poisson framework while retaining an evaluable joint likelihood and the associated computational advantages.

Background

The proposed framework uses Gaussian factor copulas to parameterize checkerboard weights for latent-intensity dependence and, where applicable, structural-zero dependence. This factor representation reduces the dimensionality of likelihood integration from the number of responses to the number of latent factors, making estimation more feasible for moderate- and high-dimensional data.

The paper notes that the framework is not restricted to Gaussian factor models, but alternative factor copula constructions can be incorporated only if their corresponding joint likelihood for the observed count vector can be evaluated. Developing such alternatives is therefore an unresolved methodological task aimed at permitting more flexible dependence structures without sacrificing computational efficiency.