Assessing the risks for natural catastrophes in property (re) insurance continues to be notably challenging for underwriters. This is due to the low frequency and high severity nature of catastrophic events. By using statistical models with high precision and strong predictive capabilities - insurers can accurately assess the varying risks found in insurance data. This allows them to measure their potential losses with confidence. Currently, the practical use of available predictive models often does not reflect these conditions and lacks the flexibility required. To address these issues, in this study, we use real-life insurance pricing data and propose models to predict and assess the average rate of loss - in the Caribbean. Specifically, our modelling approach starts by assessing the classical linear models and then extending to the linear mixed and generalized linear models. Here we focus on log-normal and gamma models to accurately capture the high severity of large losses. This study aims to propose the best suitable model given the underlying data, which attains high predictive accuracy. Our results show that the linear mixed model can predict the average loss ratio with high precision. These multilevel models account for varying effects found in different risk class levels (random effects), with fixed and interaction effects of various risk factors. We confirmed these findings by evaluating the model’s performance with new unseen (test) data sets.     «

Assessing the risks for natural catastrophes in property (re) insurance continues to be notably challenging for underwriters. This is due to the low frequency and high severity nature of catastrophic events. By using statistical models with high precision and strong predictive capabilities - insurers can accurately assess the varying risks found in insurance data. This allows them to measure their potential losses with confidence. Currently, the practical use of available predictive models often...     »