Robust nonparametric estimation of the conditional tail dependence coefficient (Q2181722)

Consider a bivariate random vector \(\left( Y^{\left( 1\right)},Y^{\left( 2\right) }\right) \) observed along with a random covariate \(X \in \mathbb{R}^{d}\). The covariate \(X\) has density function with support \(S_{X} \subset \mathbb{R}^{d}\), having non-empty interior. The continuous conditional marginal distribution functions of \(Y^{\left( j\right) }\) given \(X=x\) are denoted by \(F_j(\cdot|x)\), \(j=1,2\). Suppose that the joint conditional distribution function of \(\left( Y^{\left( 1\right)},Y^{\left( 2\right) }\right) \) satisfies that for all \(x \in S_{X}\) and \(y \in \left[ 0,1\right] \) \[ \begin{multlined} \operatorname{Pr}\left[1-F_1 \left(Y^{(1)} | x\right) < y, 1 - F_{2}\left( Y^{(2)} |x \right) < y | X=x \right]\\ = C(x) y^{\frac{1}{\eta(x)}} \left\{1+ \frac{1}{\eta(x)} \delta (y|x) \right\}, \end{multlined} \] where \(\eta\left( x\right) \in (0,1] \) is \emph{the conditional tail dependence coefficient}, and \(\left|\delta\left(\cdot|x\right) \right|\) is a regularly varying function in the neighborhood of zero. In this paper, the authors deal with a nonparametric estimation of the conditional tail dependence coefficient \(\eta\left( x\right)\) in the robust way. From the authors' abstract: The estimator is obtained by fitting the extended Pareto distribution locally to properly transformed bivariate observations using the minimum density power divergence criterion. We establish convergence in probability and asymptotic normality of the proposed estimator under some regularity conditions. The finite sample performance is evaluated with a small simulation experiment, and the practical applicability of the method is illustrated on a real dataset of air pollution measurements.