Modeling Dependencies in Claims Reserving with GEE

Abstract: A common approach to the claims reserving problem is based on generalized linear models (GLM). Within this framework, the claims in different origin and development years are assumed to be independent variables. If this assumption is violated, the classical techniques may provide incorrect predictions of the claims reserves or even misleading estimates of the prediction error. In this article, the application of generalized estimating equations (GEE) for estimation of the claims reserves is shown. Claim triangles are handled as panel data, where claim amounts within the same accident year are dependent. Since the GEE allow to incorporate dependencies, various correlation structures are introduced and some practical recommendations are given. Model selection criteria within the GEE reserving method are proposed. Moreover, an estimate for the mean square error of prediction for the claims reserves is derived in a nonstandard way and its advantages are discussed. Real data examples are provided as an illustration of the potential benefits of the presented approach.