Probabilistic loss reserving prediction via denoising diffusion model

This paper introduces an innovative approach to predicting loss reserves in the insurance industry through a revised diffusion model. This model leverages run-off triangles of claim data as graphical representations, highlighting the interconnections among data points within the triangle. Unlike the traditional cross-classified over-dispersed Poisson (ccODP) model, our proposed diffusion model not only enhances accuracy and efficiency but also provides probabilistic forecasts. Through comprehensive simulation and empirical studies, we demonstrate the superior forecasting capabilities of our diffusion model compared to existing methods. These findings indicate that using network-based interactions within run-off triangles can significantly improve loss reserve forecasting.