Sequential Hypothetical Testing for Derivative Models

As George Box [1976] put it, "all models are wrong but some are useful." This is especially true for derivative models, As sown during the 2008 financial crisis, most collaterized debt obligation (CDO) models failed to reflect the true risk of those structures. Therefore, one always faces the question "which model best describes the market dynamics given the data at hand" when using models for derivative pricing and risk management. Traditionally, the evaluation of model usage relies on the judgements of model users and reviewers, which could be subjective because it heavily relies on their backgrounds, experience and incentives. To address this issue, we propose a statistical framework to objectively assess the performance of derivative models by evaluating the outcome of using certain hedging strategy. We also performed type two error analysis based on this framework. This is useful in identifying potential risk from wrongly accepted model usage