ROCKET: robust confidence intervals via Kendall's tau for transelliptical graphical models (Q1990586)

The authors consider the problem to estimate entries in the precision matrix \(\Omega\) of a transelliptical distribution on a random graph \(G\). Observations are \(n\) independent samples of a transelliptic distribution with correlation matrix \(\Sigma = \Omega^{-1}\) and functions \(f_1, \dots, f_p\). Fix two indices \(a\) and \(b\) and consider the entry \(\Omega_{a,b}\). To estimate this entry, initial estimates of the regression coefficients of the marginal transform \(f_a \left(X_a\right), f_b\left(X_b\right)\) on the remaining marginals are used. Those are then combined with an estimator \(\hat \Sigma\) of \(\Sigma\) based on Kendall's tau. For fixed indices \(a\) and \(b\), \(\sqrt{n}\)-consistency and asymptotic normality of the investigated estimator are proven, and the latter also allows to construct an asymptotic confidence interval for the true entry. The finite sample performance of the method is illustrated in a simulation study.