scientific article; zbMATH DE number 17212

From MaRDI portal
Publication:3973909





[For the entire collection see Zbl 0722.00032.] The authors consider the problem of nonparametric estimation of the d- dimensional elliptically contoured density \[ f(x)=(\text{det }\Sigma)^{-1/2}k(x^ t\Sigma^{-1}x), \] where \(x^ t\) represents the transpose of the vector \(x\). Both the positive definite symmetric matrix \(\Sigma\) and the transfer function \(k\) are unknown. Let \(\Sigma_ m\) be an estimate of \(\Sigma\) based on an independent random sample of size \(m\), and let \(X_ 1,\dots,X_ n\) be a random sample. A kernel density estimate of the form \[ f_ n(x)=\{cnh_ n\}^{-1}\{x^ t\Sigma_ m^{-1}x\}^{-(d-2)/2}\sum_{i=1}^ n K((x^ t\Sigma_ m^{-1}x- X^ t_ i\Sigma_ m^{-1}X_ i)/h_ n), \] where \(h_ n\) is the bandwidth and \(K\) is the kernel, is proposed. Under suitable conditions on \(h_ n\), \(K\) and \(\Sigma_ m\), asymptotic normality of \(f_ n\) is established. The interesting fact here is that the convergence rate is free of \(d\). As anticipated, simulation results show that \(f_ n\) outperforms the \(d\)-dimensional classical kernel estimate quite well when \(f\) is assumed to be elliptically contoured.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3973909)