We propose a semiparametric approach called elliptical skew-(S)KEPTIC for efficiently and robustly estimating non-Gaussian graphical models. Relaxing the assumption of semiparametric elliptical distributions to the family of \textit{meta skew-elliptical} that accommodates a skewness component, we derive a new estimator which is an extension of the SKEPTIC estimator in Liu et al. (2012), based on semiparametric Gaussian copula graphical models, to the case of skew-elliptical copula graphical models. Theoretically, we demonstrate that the elliptical skew-(S)KEPTIC estimator achieves robust parametric convergence rates in both graph recovery and parameters estimation. We conduct numerical simulations to prove the reliable graph recovery performance of the elliptical skew-(S)KEPTIC estimator. Finally, the new method is applied to the daily log-returns of the stocks of the S\&P500 index and shows better interpretability compared to the Gaussian copula graphical models.
翻译:暂无翻译