Greed, lack of transparency and insufficient regulatory oversight

Almost everybody agrees that these were the sources of the current financial crisis. A fundamental question is whether the decision support, the risk modeling or simply the management tools portrayed the risk properly or not. By the way, properly does not mean exactly – nobody has a crystal ball. Many board members would have met behind locked doors, even if their risk managers warnings would have been an order of magnitude below the current losses.

In other words: Did the risk managers roughly know what could happen in an extreme event and did not want to say it because unless it happens they and their company made an excessive amount of money. Or, were they clueless because the loss scenarios produced by their models were so far off? In the first case they should be fired because they cheated all stakeholders. The latter case is more delicate.

How much should we believe in the modeling results? First of all, the models are only one thing. Their calibration is another very important issue. The products and the market place have become too complex that a single person could master all the calibration and modeling issues that arise in estimating a firms loss potential.

If one person cannot grasp it, maybe a group could or even a whole community? Preferably the group or the community should be heterogeneous; members of the same corporate culture usually have the same blind spots. But how could a broad community discuss, understand or even improve the models if most of them come from proprietary vendors? Jaikumar Vijayan discusses some of these issues in his article Needed: Better IT crystal balls on Wall Street.

I believe there is a simple solution: Use open source components and models. The communication about the models and the results immediately gets raised to a higher level because all participating partners (including regulators) have complete access to the risk management models – they form the community which is required to understand the models' results. Isn't it odd that normally, this community has to bang at closed doors of proprietary software vendors who claim that their, usually small group of experts, have chosen the best methods and you should trust them. I certainly feel more comfortable if I can study and run test cases, compare results and if necessary also implement my own algorithms.

Markus