Abstract
Claim reserving is primarily accomplished using macro-level or aggregate models, with the Chain-Ladder method being the most popular one. However, these methods are heuristically constructed, rely on oversimplified data assumptions, neglect the heterogeneity of policyholders, and so lead to a lack of accuracy. In contrast, micro-level reserving leverages on stochastic modeling with granular information for improved predictions, but usually comes at the cost of more complex models that are unattractive to practitioners. In this paper, we introduce a simplistic macro-level type approach that can incorporate granular information at the individual level. We do so by considering a novel framework in which we view the claim reserving problem as a population sampling problem and propose an estimator using inverse probability weighting techniques, with weights driven by policyholder attributes. The framework provides a statistically sound method for aggregate claim reserving in a frequency and severity distribution-free fashion, while also incorporating the capability to utilize granular information via a regression-type framework. The resulting reserve estimator has the attractiveness of resembling the Chain-Ladder claim development principle but applied at the individual claim level, so it is easy to interpret by actuaries, and more appealing to practitioners as an extension of a macro-model.
