I think we would benefit from better and more realistic models of systems in economics, in finance, better models for assessing risks, and so on. But having a netter model is on thing, using it is another entirely.he wrote.
It seems risk models in finance are repeatedly abused to suite the need to higher ups.
Emanuel Derman and Paul Wilmott, publishing their Financial Modeler's Manifesto, kind of suggested we don't need more complicated models 2008.
There are behavioral and technical aspects in this topic. In my Review of the Blame Games of 2008 I have touched some issues including VaR that has been used to hide risk instead of exposing it.
In VaR is Dead - Long Live VaR I referred to the investigations concerning the "London whale trade" an obvious abuse ... Ironically enough the US Senate Subcommittee recommended ... drumroll ... more complex models. But it seems more complex models allow more manipulations. This is a bit more then the insight that we all cheat a little (Dan Ariely). But there are enough people touching the moral aspects ...
The risky horror has also a few technical sides:
What is the use of more information (factors) if you lose them in the numerical jungle? - poor solvers or calibration engines that produce horror oscillations....With the recommendation: stop when good enuf is great.
What if you apply a great but wrong model to a great instrument? - in the picture you see a simple Reverse Floater valued by HW versus a Black Karasinski models over a longer time period. Non-callable the fair value is model-invariant, but callability makes it model dependent. There is model risk that cannot be hedged away and you need to perform a lot of across-mode scenarios.
If you complicate the models you might fall into the trap above at each single valuation and leopards, poisonous snakes and spiders attack you in the jungle ...
Use more realistic instrument-model combinations and models that are adequate and applicable.
Having said this, I am interested how the bilateral CVA/DVA ... in pricing a single instrument requirements discussion will "continue".
We have the right math and also the implementation techniques, but ....
