A1077
Title: Probabilistic loss reserving prediction via denoising diffusion model
Authors: Shiying Gao - The University of Sydney (Australia) [presenting]
Boris Choy - University of Sydney (Australia)
Yuning Zhang - The University of Sydney Business School (Australia)
Junbin Gao - The University of Sydney (Australia)
Ruikun Li - The Univeristy of Sydney (Australia)
Abstract: The aim is to propose a novel approach for predicting loss reserves within the insurance industry using a revised diffusion model. This approach considers the run-off triangles of claim data as graphical representations to elucidate connections among data points within the triangle. In contrast to the traditional cross-classified over-dispersed Poisson (ccODP) model, the diffusion model not only exhibits enhanced accuracy and efficiency but also provides probabilistic forecasts. Through a thorough analysis encompassing both simulation and empirical studies, the superior forecast accuracy of the diffusion model is demonstrated compared to existing methodologies. These results suggest that harnessing network-based interactions within run-off triangles holds promise for improving loss reserve forecasting.
