Forward likelihood-based predictive approach for space–time point processes

Dealing with data from a space–time point process, the estimation of the conditional intensity function is a crucial issue even if a complete definition of a parametric model is not available. In particular, in case of exploratory contexts or if we want to assess the adequacy of a specific parametric model, some kind of nonparametric estimation procedure could be useful. Often, for these purposes kernel estimators are used and the estimation of the intensity function depends on the estimation of bandwidth parameters. In some fields, like for instance the seismological one, predictive properties of the estimated intensity function are pursued. Since a direct ML approach cannot be used, we propose an estimation procedure based on the subsequent increments of likelihood obtained adding an observation one at a time. Simulated results and some applications to statistical seismology are provided. Copyright 2011 John Wiley & Sons, Ltd.