Sébastien Jessup | Searcharxiv

arXiv2026

A zero-inflated mixed-effects spatial point process for grouped storm loss data

The increasing granularity of third-party weather and exposure information can allow insurers to more effectively predict weather-related losses. However, loss outcomes are often reported in spatially grouped observations, such as at the county level, so higher resolution predictors are aggregated to align with the granularity of the outcome in standard analyses. Assuming an underlying zero-inflated mixed-effects spatial point process framework for claims arising from a common storm, we derive a model for unbalanced, multivariate zero-inflated count data that incorporates rich weather and exposure predictors observed at higher spatial granularity to predict claim patterns. The model accommodates the dependence between locations affected by a common storm in the excess zeros, as well as in the joint claim counts. Using real property exposure and loss data, we emphasize the value of incorporating granular predictors to address the localized heterogeneity of storm losses.

stat.AP↗

arXiv2025

Flexible extreme thresholds through generalised Bayesian model averaging

Insurance products frequently cover significant claims arising from a variety of sources. To model losses from these products accurately, actuarial models must account for high-severity claims. A widely used strategy is to apply a mixture model, fitting one distribution to losses below a given threshold and modeling excess losses using extreme value theory. However, selecting an appropriate threshold remains an open question with no universally agreed-upon solution. Bayesian Model Averaging (BMA) provides a promising alternative by enabling the simultaneous consideration of multiple thresholds. In this paper, we show that an error integration BMA algorithm can effectively detect heterogeneous optimal thresholds that adapt to predictive variables through the combination of mixture models. This method enhances model accuracy by capturing the full loss distribution and lessening sensitivity to threshold choice. We validate the proposed approach using simulation studies and an application to an automobile claims dataset from a Canadian insurer. As a special case, we also study the homogeneous setting, where a single optimal threshold is selected, and compare it to automatic selection algorithms based on goodness-of-fit tests applied to an actuarial dataset.

stat.ME↗