Bayesian Forecasting for Accident Proneness Evaluation (Q4512138)

Insurance companies are interested in questions concerning number of accidents and accident proneness of their insurants, to forecast how much they should expect to pay or which premiums should they charge. Given the enormous socioeconomic implications of labour accidents, evaluating accident proneness of companies should also help to identify where to allocate resources to improve safety at the workplace. This is recognized in several countries, including Spain, where, by law, insurance companies need to spend part of their budgets in that task, typically through educational campaigns. Answers to all these questions demand appropriate forecasting systems. This paper describes such a system, stemming from previous work of some of the authors with MAPFRE, a Spanish insurance company. The general underlying theme is Bayesian forecasting. Bayesian methods are specially fit for that task since they allow for the coherent incorporation of all sources of information, including expert opinions and data. Three increasingly complex and realistic models for that purpose are described in the paper. For predictive and inference purposes, authors rely on simulation methods. The model with a real case is illustrated and their implementation in a forecasting system developed for an insurance company is described.