Claims reserving and generalised additive models

This paper shows how non-parametric smoothing can be applied in the context of claims reserving. The paper concentrates on the chain-ladder technique, within the framework of the chain-ladder linear model, but the methods can easily be applied to other models. It is shown that non-parametric smoothing can provide more stable reserve estimates, and is an alternative to other methods suggested for this purpose such as the Kalman filter. The methods are implemented in the statistical package S-PLUS.