Bayesian Multivariate Nonlinear State Space Copula Models

A novel flexible class of multivariate nonlinear non-Gaussian state space models, based on copulas, is proposed. Specifically, it assumed that the observation equation and are defined by copula families not necessarily equal. Inference performed within Bayesian framework, using Hamiltonian Monte Carlo method. Simulation studies show proposed copula-based approach extremely flexible, since able to describe a wide range dependence structures and, at same time, allows us deal with missing data. The application atmospheric pollutant measurement data shows suitable for accurate modeling prediction dynamics in presence values. Comparison Gaussian linear model additive regression trees superior performance respect predictive accuracy.