The validation of derivative pricing models can be a slow, labor intensive and expensive exercise. Moreover, it often provides a limited amount of certainty on the correctness of the pricing models. However, since pricing models are mathematical models, they satisfy mathematical identities which can provide strong tests that leave very little possibility for error. Furthermore, these tests provide failure conditions that require no human judgment, that can be automated, and that can therefore run over tens of thousands of test scenarios.
In this paper, David Eliezer, PhD, Vice President and Head of Model Validation at Numerix, explores new approaches practitioners can utilize to improve their model validation processes.
Highlights include:
- Overview of model validation as it is practiced today
- Parallel implementation, convergence and stress testing
- Introduction to testing with mathematical identities
- How mathematical identities of models can be used specifically
- PDE testing
- Calibration round trip testing
- Financial correctness and hedge performance
