zbMATH — the first resource for mathematics

A kernel method for smoothing point process data. (English) Zbl 0584.62140
Let \((x_ 1,...,x_ n)\) be a sample of a point process observed on (0,T). The author considers smoothing of these data by a kernel in the following way \[ {\hat \lambda}_ t(x)=\sum^{n}_{i=1}\delta (| x-x_ i| /t)/\int^{T}_{0}\delta (| x-u| /t)du \] where \(\delta\) is a symmetric density. In the case of a Cox process \({\hat \lambda}{}_ t(x)\) is an estimate of the underlying realization of the rate process. It is not quite clear what \({\hat \lambda}{}_ t(x)\) estimates for a general point process, but it can serve as a useful tool for the explanatory analysis of the heterogeneity of an observed process.
The paper gives a procedure for the selection of the bandwidth t. In the case of a Cox process the mean-square-error \(M(t)=E(({\hat \lambda}_ t(x)-\lambda (x))^ 2)\) can be expressed for uniform \(\delta\) in terms of the intensity \(\mu\) and the reduced second moment measure K. The procedure selects t by minimizing \(\hat M(\)t) obtained from M(t) by substituting the usual estimators \({\hat \mu}\) and \(\hat K\) for \(\mu\) and K. The method is illustrated by simulated data and some real data on joints along a coal seam.
Reviewer: H.R.K√ľnsch

62M09 Non-Markovian processes: estimation
62G05 Nonparametric estimation
93E14 Data smoothing in stochastic control theory
60G55 Point processes (e.g., Poisson, Cox, Hawkes processes)
62M20 Inference from stochastic processes and prediction
Full Text: DOI