Asymptotic consistency of fixed-width sequential confidence intervals for a multiple regression function (Q1073520)

Let \(m_ n(x)\) be the recursive kernel estimator of the multiple regression function \(m(x)=E[Y| X=x]\). For given \(\alpha\) \((0<\alpha <1)\) and \(d>0\) we define a certain class of stopping times \(N=N(\alpha,d,x)\) and take \(I_{N,d}(x)=[m_ N(x)-d\), \(m_ N(x)+d]\) as a 2d-width confidence interval for m(x) at a given point x. In this paper it is shown that the probability \(P\{m(x)\in I_{N,d}(x)\}\) converges to \(\alpha\) as d tends to zero.