In this presentaon we consider regression based reserving models that allow for separate RBNS and IBNR reserves based on aggregated discrete me data containing informaon for different combinaons of accident years, reporng delay, and payment delay. The models that are analysed either describe both claim count and payment amount dynamics, or only payment amount dynamics. All introduced models will be closely related to the cross-classified over-dispersed Poisson (ODP) chain-ladder model.

Further, these general ODP models will be esmated using regression funcons defined by tree-based gradient boosting machines (GBM). This will provide us with machine learning based reserving models that have interpretable output, and that are easy to bootstrap from. We will give a brief introducon to GBMs, including basic calibraon and model selecon, and illustrate the reserve performance based on complex simulated data. The presentaon is based on joint work with M. Lindholm, R. Verrall, and F. Wahl.