Fitting a multiple regression function (Q792724): Difference between revisions

The authors study the problem of nonparametric estimation of an unknown function \(g(\cdot)\) in the multiple regression model \(Y_ j^{(n)}=g(x^ n_ j)+e_ j^{(n)}\) where \(x_ j^{(n)}\) are known points in the p-dimensional unit cube, \(Y_ j^{(n)}\), 1\(\leq j\leq n\), are the observables and \(\{e_ j^{(n)}\}\) are i.i.d. random variables with mean 0 and finite variance \(\sigma^ 2\). They propose a kernel type estimator for \(g_ n(x)\) based on a known p-dimensional bounded density \(k(\cdot)\) and study its asymptotic properties. They also obtain a consistent estimator of \(\sigma^ 2\). Optimum choice of the kernel \(k(\cdot)\) in the sense of minimizing the mean square error of the estimator \(g_ n(x)\) is discussed.