scientific article; zbMATH DE number 3936216

From MaRDI portal
Publication:3707124





The author studies the rate of mean-square convergence of the predictogram estimator of the conditional expectation \(R(X_{n- k+1},...,X_ n)\equiv E\{g(X_{n+s})| X_{n-k+1},...,X_ n\}\) based on a known function \(g:[0,1]^ p\to {\mathbb{R}}\) and on observations \(\{X_ 1,...,X_ n\}\) from a strictly stationary and \(\phi\)-mixing time series \(\{X_ i\}^{\infty}_{i=1}\), as the sample-size parameter n goes to \(\infty.\) The observations \(X_ i\) are assumed to lie in \([0,1]^ p\), and a sequence \({\mathcal P}_ n\) of partitions of \([0,1]^{kp}\) is fixed in defining the predictogram \[ R_ n(X_{n-k+1},...,X_ n)\equiv \sum^{n-s}_{i=k}g(X_{i+s})I_{[(X_{i-k+1},...,X_ i)\in J]}/\sum^{n-s}_{i=k}I_{[(X_{i-k+1},...,X_ i)\in J]} \] where J is the element of \({\mathcal P}_ n\) containing \((X_{n-k+1},...,X_ n).\) The mean-square difference \(R_ n(X_{n-k+1},...,X_ n)-R(X_{n- k+1},...,X_ n)\) is bounded in terms of bounds on g, of the mesh-size of \({\mathcal P}_ n\) and of Lipschitz constants for R(\(\cdot)\).











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 Q3707124)