scientific article; zbMATH DE number 3958451

From MaRDI portal
Publication:3727142





The author uses a robust method introduced by \textit{P. J. Huber} [Robust statistics. (1981; Zbl 0536.62025)], for solving problems with regard to independent random variables. The purpose of this paper concerns the possibility of defining a dependence measure of type R (estimates derived from rank tests). The author studies two classes of robust measures of linear dependence for stationary stochastic processes. The first class is based on one- sample rank tests, the second one on rank correlation coefficients. First, the author utilises the nonrobustness of the autocorrelation coefficient and presents general characteristics of the considered measures. Let \(\{Y_ i\), \(i\in {\mathbb{Z}}\}\) be a real stationary stochastic process. Suppose that only a part \(\{y_ 1,y_ 2,...,y_ n\}\) of this process is observed. The problem is to define a statistic \(\theta^{(n)}\), which is an indicator of the linear dependence of \(Y_ i\) and \(Y_{i+k}\) and is robust against outlier observations [\textit{F. R. Hampel}, Ann. Math. Stat. 42, 1887-1896 (1971; Zbl 0229.62041)]. The paper contains some asymptotic properties of the considered measures and discusses certain connections between these and the autocorrelation function [\textit{H. Bergström}, Weak convergence of measures. (1982; Zbl 0538.28003) pp. 82]. Monte Carlo small sample results are exhibited, too.












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 Q3727142)