Multivariate local fitting with general basic functions (Q1887216)

This paper demonstrated how useful concepts such as multivariate local smoothing and fitting with general basis functions can be combined to improve the estimation properties of smooth fits. The author starts with arbitrary basis functions and studies the asymptotic variance both at interior and boundary points. Then, in the case of basic functions without interactions, he proves a new Taylor-type theorem which provides asymptotic expressions for bias and variance of the corresponding local estimator. A data-driven tool to obtain a suitable basis is also presented. The performance of fitting procedures with general basic functions are demonstrated via an example with a simulated data set.