A Bayesian Approach for Computing Claim Amounts of Occuring but Not Reported Events

Abstract

For many reasons, insurance companies often do not report outstanding claims as soon as they occur (incurred but not reported or IBNR). Instead, there is a time interval between the time of occurrence and the claims. Therefore, in practice it is of central interest to provide reserves for these outstanding claims. In this project, several Bayesian models are proposed and compared in terms of their predictive capability. Inference for these models is not trivial and Markov chain Monte Carlo (MCMC) methods are used to estimate and compare models. We are currently using packages WinBUGS and JAGS which are freely available.

Participants

  1. Leonardo Melo

Some references

[2002, article]
Ntzoufras, I., & Dellaportas, D. (2002). Bayesian Modelling of Outstanding Liabilities Incorporating Claim Count Uncertainty (with discussion). North American Actuarial Journal, 6(1), 113-128.
[1998, article]
Renshaw, A., & Verrall, R. (1998). A Stochastic model Underlying the Chain-Ladder Technique. British Actuarial Journal, 4, 903-926.