Similarity measures of conditional intensity functions to test separability in multidimensional point processes

Separability in the context of multidimensional point processes assumes a multiplicative form for the conditional intensity function. This hypothesis is especially convenient since each component of a separable process may be modeled and estimated individually, and this greatly facilitates model building, fitting, and assessment. This is also related to the problem of reduction in the number of dimensions. Following previous approximations to this problem, we focus on the conditional intensity function, by considering nonparametric kernel-based estimators. Our approach calculates thinning probabilities under the conditions of separability and nonseparability and compares them through divergence measures. Based on Monte Carlo experiments, we approximate the statistical properties of our tests under a variety of practical scenarios. An application on modeling the spatio-temporal first-order intensity of forest fires is also developed.