Bayes and Empirical Bayes Estimation for the Chain Ladder Model

Many authors have criticized the chain ladder, or "development factor," model as over-parameterized. This paper does not deal with this criticism but does address other areas of concern for practitioners using the model. Verrall approaches the problem of estimating the parameters of the linear (after log-transform) chain ladder model from a Bayesian perspective. Of course, without prior information, the Bayesian estimates are simply the maximum likelihood parameter estimates. With prior information available, Verrall uses Bayesian and empirical Bayesian methods to incorporate both the observed data and the prior information in "credibility-like" formulae to derive parameter estimates with less standard error. This could be useful for those practitioners who deal with small data sets and need to augment those data with other experience. It also provides a smooth way for adapting to changes in underlying patterns, to the extent they are credible.