Baldly, you only see what you're looking for.  

It's not that bad. When you have a model (produced by scratching you head), and you miss a few calibration parameters, extracting them from the market is like asking around a closed question (answer: yes/no or a single digit, or which box gets checked). In many cases, this actually works very well, pretty safe, and anyway there's no other way you can cope with the volume dealt everyday (I'm thinking of plain vanilla put/call options here, on stocks or currencies).

Like every closed question, it has a preconception of the world. The answer will not tell you that may be you should take a broader perspective and rethink your model because times are changing. You only realize when the results become unstable with time, or you lose money... And then you make the model smarter (like Local Volatility vs. Black-Scholes after the 1987 crash)

The problem is credit derivatives have had no prior "model testing, low volume phase". The bubble brought the volumes to the top tier of the banking business in 5 years.

Pierre