Bootstrap-based Significance Tests for Poisson Autoregression Models with an Application to Extra-Tropical Cyclones (joint with Andreea Halunga)

Abstract:  Estimation and testing on the boundary of the parameter space is nontrivial, particularly if the variable of interest is discrete. We derive the asymptotic distribution of the quasi-maximum likelihood estimator asymptotic for the Poisson autoregression model with exogenous covariates. We then develop asymptotic and parametric bootstrap Wald tests for the significance of the models parameters on the boundary of the parameter space. We apply these results to test for temporal clustering in extra-tropical cyclones and show that storm clustering is driven not only by large-scale atmospheric conditions but also by means of secondary cyclogenesis (a parent cyclone that generates offsprings).