Backtesting the Bayesian Bornhuetter-Ferguson method against traditional approaches in claims reserving

Abstract

Evaluation of claims reserve is a paramount process for non-life insurance company. To this end, several deterministic and stochastic methodologies have been provided in the literature. Therefore, the validation of the models on actual data and the comparison of these models appropriateness is nowadays a crucial question. We focus here on different Bornhuetter-Ferguson methodologies and we backtest the behavior of these models using the well-known dataset made available in [22]. The aim is to test both the ability of different models to well predict future losses as well as to evaluate the effects of different priors on the results. Additionally, we test the uncertainty of the predictions by comparing the coefficient of variation.

Subject Classification:

• 62P05

Keywords:

• Claims reserves
• Bornhuetter-Ferguson methodologies
• Backtesting
• Markov-Chain-Monte Carlo