Modeling the occurrence of vehicle claims for the State of Minas Gerais - Brazil via Bayesian inference

Abstract

Claims occurrence modeling is commonly done using logistic regression, but this model can present low predictive capacity in situations of unbalanced data. Alternatively, regression models that deal with excess zeros, such as zero-inflated or zero-adjusted distributions, can be used. In this paper, we aim to model the occurrence of claims that resulted in total loss in vehicles in the state of Minas Gerais in 2019. We considered the Zero Adjusted Binomial (ZABI) regression with the Bayesian framework, inserting covariates related to the characteristics of the insured. We noted that risk profiles of policyholders characterized by the variables gender and age were significantly associated with the claim, suggesting a reduction on the probability of occurrence in female policyholders, when compared to male policyholders. Under the Bayesian approach, this model allows the insertion of the actuary's prior knowledge in relation to the analyzed event using informative priors.