Model Validation - First Defense Front of Risk Management?

Computational Finance is, in our understanding, the art of modelling financial processes and then to somehow calculate fair values of financial instruments. 

The continuing financial crisis has led to significant changes in the valuation of derivative contracts. New accounting standards force the adjustment of values with respect to many risk factors and introduce much more complexity to the valuations space - and consequently traps that should be avoided.


About 3 years ago we have presented traps that are hidden in model complexity and poor numerical methods, especially in the model calibration - like, in the Risky Horror Show in the frame of the Wilmott Finance Focus in London.

This is not only important for the valuation of exotic instruments, but also for standard pricing of vanilla instruments - think of the model traps when interest rates can become negative. In the The Worst Kind of Pricer I referred to an experience of Stefan Fink of Raiffeisen Upper Austria, who calls some vanilla instruments the "new structured ones" recommending multi-model valuations with UnRisk.


But multi-model valuation is not the only answer. Multiple sources for prices are often disagreeing with one another and it is difficult to find patterns why and how.


Better risk management is not only focussed on portfolio level, but in the specific instrument valuation. At the other hand portfolio valuation (including exposure modeling) influences the fair value of an instrument - in general, valuation requires information that is usually extracted "later".


Most process industries have achieved inline quality management - quality is not measured at the finished products, but determined by controlling the process behavior towards quality - this includes the validation and adaptation of the process models, ...


In the analogy, derivative model validation would be part of the control process working towards quality goals of financial instruments (represented by models) related to their purpose. Quality means fit for purpose, accuracy and robustness.


Ironically that requires billions of valuations and only blazingly fast engines that can value the scenario bundles necessary to detect and avoid "defects". 

I am not sure, whether all market participants, regulators and technology providers have screened the full consequences with respect to mathematical finance, heuristics and data driven methodology?