Challenges with non-informative gamma priors in the Bayesian over-dispersed Poisson reserving model

Tables 2 and 3 in England et al. (2012) raise the conjecture that the claims reserves in the Bayesian over-dispersed Poisson (BODP) model with non-informative gamma priors are equal to the claims reserves in the chain-ladder (CL) model (the small differences in the figures could be explained by simulation inaccuracy). We give a counter-example to this conjecture. Moreover, we show that, working with non-informative priors needs a lot of care because a naïve treatment may lead to improper posterior distributions.