The stable tail dependence function in \(x\) can be estimated by
$$ \hat{l}_k(x) = 1/k \sum_{i=1}^n 1_{\{\exists j\in\{1,\ldots, d\}: \hat{F}_j(X_{i,j})>1-k/n x_j\}}$$
with
$$\hat{F}_j(X_{i,j})=(R_{i,j}-\alpha)/n$$
where \(R_{i,j}\) is the rank of \(X_{i,j}\) among the \(n\) observations in the \(j\)-th dimension:
$$R_{i,j}=\sum_{m=1}^n 1_{\{X_{m,j}\le X_{i,j}\}}.$$
This estimator is implemented in stdf
.
The second estimator is given by
$$ \tilde{l}_k(x) = 1/k \sum_{i=1}^n 1_{\{X_{i,1}\ge X^{(1)}_{n-[kx_1]+1,n} or \ldots or X_{i,d}\ge X^{(d)}_{n-[kx_d]+1,n}\}}$$
where \(X_{i,n}^{(j)}\) is the \(i\)-th smallest observation in the \(j\)-th dimension.
This estimator is implemented in stdf2
.
See Section 4.5 of Beirlant et al. (2016) for more details.