A hierarchical reserving model for reported non-life insurance claims

Traditional non-life reserving models largely neglect the vast amount of information collected over lifetime a claim. This includes covariates describing policy, claim cause as well detailed history during claim's development time. We present hierarchical model modular framework for integrating and claim-specific into process. Hierarchical decompose joint likelihood process Moreover, they are tailored to portfolio at hand by adding layer each events registered (e.g. settlement, payment). Layers modelled with statistical learning generalized linear models) or machine methods gradient boosting machines) use covariates. As result its flexibility, this incorporates many existing models, ranging from aggregate designed run-off triangles individual using connection allows us develop data-driven strategy choosing between reserving; an important decision practitioners. illustrate our method case study on real insurance data set deduce new insights in driving claims. we evaluate method's performance large number simulated portfolios representing several realistic scenarios demonstrate flexibility robustness model.