Empirical Method-Based Aggregate Loss Distributions

This paper presents a methodology for constructing a deterministic approximation to the distribution of the outputs produced by the loss development method (also known as the chain-ladder method). The approximation distribution produced by this methodology is designed to meet a preset error tolerance condition. More specifically, each output of the loss development method, when compared to its corresponding approximation, meets the preset error tolerance. Ways to extend this methodology to the Bornhuetter-Ferguson and the Berquist-Sherman families of methods are described. The methodology is illustrated for a sample loss development history.