No label defined (Q4259474)

Let \(F(x) (x\in\mathbb R)\) be a distribution function and \(\{x_i\}\) be the observation points, \(a=x_0<x_1<\ldots<x_n=b\), where \(-\infty<a<b<\infty\). Restoration of \(F(x)\) by the values \(Y_i=F(x_i)+\varepsilon_i,\;i=0,\ldots,n\), where \(\varepsilon_i\) are random errors is considered. It is supposed that the errors \(\varepsilon_i\) are independent and \(M\varepsilon_i=0, D\varepsilon_i=\sigma^2<\infty\) for every \(i\), the points \(x_i\) are such that \(r_n=\max_{1\leq\;i\leq n}r_i\), where \(r_i=x_i-x_{i-1}\). As an estimator of \(F(x)\) the following statistic is considered: \[ \hat F_n(x)=h^{-1}(n)\sum_{i=1}^nY_iK((x-x_i)/h(n))(x_i-x_{i-1}), \] \(K(x)\) is a function with bounded derivative on \([a,b]\) satisfying the condition \(\int_{-\infty}^\infty K(x) dx=1\), \(h(n)>0\) and \(h(n)\to 0\) for \(n\to\infty\). Under some additional conditions on \(h(n)\) and \(F(x)\) the estimator \(\hat F_n(x)\) is an asymptotically unbiased and consistent estimate for the function \(F(x)\).