Variance in claim reserving

Document description

This paper is educational. There has been considerable development of statistical techniques for predicting claim payments over recent years, which has yet to be assimilated by the profession and put into practice by practitioners. We want to spread the knowledge of these techniques, to dispel some of the mystique, and to give some examples which demonstrate how they work in practice. These techniques do not replace existing methodologies, but serve to enrich the actuary's tool box. So that readers can form a view of the success of existing methods in the past we include a review of the variance of the actual out-turn from the reserves of some UK insurance companies over the past ten or so years, to which we add some thoughts on the factors which may have contributed to the variances. We would like to encourage a healthy scepticism of 'black box' techniques and some of the pitfalls for the unwary are presented as a warning against using them without an understanding of the limitations. For example we believe that the use of the term 'Confidence intervals' is to be discouraged since we think it conveys a false impression of the modelling process, which applies to past data. The circumstances that will apply in the future can not be known at the moment, so the model is emphatically not a crystal ball.