We propose a simulation-based approach to compare probabilistic hazard and risk models, exploiting the Bayesian prior/posterior predictive p-values (PPP) framework. The comparison can utilize an arbitrary summary statistic and can be customized to the aspects of interest, particularly the right tail, which is crucial in risk assessment. The primary benefits of our approach in comparison to existing alternatives are twofold. Firstly, it incorporates both aleatory and epistemic variability in a natural probabilistic framework, secondly, it produces interpretable measures of discrepancy. The method is demonstrated on synthetic data and two state-of-the-art seismic hazard models for Italy (MPS19, Modello di Pericolosità€ Sismica 2019, and ESHM20, European Seismic Hazard Model 2020). The method is applicable in any domain involving probabilistic hazard or risk models, including flood, volcanic, or multi-layer single hazard or single risk assessments.
