Nonparametric Regression with Spatially Dependent Data

In this paper we present a new procedure for nonparametric regression in case of spatially dependent data. In particular, we modify the two-step local linear regression of Martins-Filho and Yao (2009) by introducing information on spatial dependence via a nonparametric estimate of the error covariance matrix. The finite sample performance of our proposed procedure is then shown via Monte Carlo simulations for various data generating processes and its practical usage is illustrated through an application to the familiar crime data set for 49 Columbus neighbourhoods.