Risk that underwriting, pricing or risk management decisions are based on models whose assumptions or calibration prove incorrect, generating systematic estimation errors.
Model risk refers to exposure to losses or decision errors resulting from a failure of the model used to represent a complex phenomenon. In the insurance and reinsurance context, it manifests primarily in three forms. Specification risk arises when the model's mathematical structure is ill-suited to the reality of the modeled phenomenon, for example a normal tail model underestimating extreme events. Calibration risk arises when the model is correctly specified but calibrated on insufficient, unrepresentative or obsolete data: this is the central problem of climate change for catastrophe models, whose historical data no longer reflect the future. Usage risk arises when the model is applied outside its validity perimeter or when its implicit assumptions are not understood by users. Concentration on a small number of model vendors (Verisk, Moody's RMS, CoreLogic) creates systemic model risk: if all market participants use the same risk representations and calibration scenarios, their portfolios share the same estimation errors. An extreme realization that challenges the dominant models then affects the entire market simultaneously, transforming an individual estimation error into a systemic shock. This phenomenon is aggravated by climatic non-stationarity, which progressively renders models calibrated on historical data obsolete.
During Hurricane Ian in 2022, catastrophe models from leading vendors underestimated insured losses by approximately 30 to 40 percent compared to actually observed losses. This systemic model error simultaneously affected insurers, reinsurers and ILS funds, which had all used similar models to price their Florida exposures and define their cat bond tranches.
model risk, risque de modèle catastrophe, incertitude de modèle