Title: Representing sparse Gaussian DAGs as sparse R-vines allowing for non-Gaussian dependence
Authors: Dominik Mueller - Technische Universitaet Muenchen (Germany) [presenting]
Claudia Czado - Technische Universitaet Muenchen (Germany)
Abstract: Modeling dependence in high dimensional systems has become an increasingly important topic. Most approaches rely on the assumption of a joint Gaussian distribution such as statistical models on directed acyclic graphs (DAGs). They are based on modeling conditional independencies and are scalable to high dimensions. In contrast, vine copula based models can accommodate more elaborate features like tail dependence and asymmetry. This flexibility comes however at the cost of exponentially increasing complexity for model selection and estimation. We show a connection between these two model classes by giving a novel representation of DAG models in terms of sparse vine copula models. Therefore, we can exploit the fast model selection and estimation of sparse DAGs while allowing for non-Gaussian dependence in the vine copula models. We evaluate our methodology by a large scale simulation studiy and high dimensional data examples demonstrating that our approach outperforms standard methods for vine copula structure selection.