Making Tweedie's compound Poisson model more accessible (Q825297)

The authors of the paper revisited the compound Poisson model with i.i.d. gamma claim sizes. This model allows for two different parametrizations, the Poisson-gamma parametrization and Tweedie's compound Poisson parametrization. The results derived illustrate when these two parametrizations are identical. The main theoretical results of the paper give conditions under which the different generalized linear model parametrizations lead to identical predictive models. These results provide a remarkable property that allows to lower calibration in Tweedie's double generalized linear models. In the applied part, authors of the paper analyze why the insurance industry gives preference to the Poisson-gamma parametrization. Based on examples, authors find that, this parametrization is easier to fit, and results turn out to be more accurate. In addition, the examples under consideration show that the Tweedie version is computationally clearly lacking behind the Poisson-gamma case.