Saturday, October 11, 2008

Is there a Compressed Sensing Solution for the Financial Regulators ?

Looking at the low traffic of this past week, I would not be surprised that everybody is very much concerned about their 401K or equivalent retirement plans and feel a little powerless about the whole situation. It's a little bit like the Titanic, everyone knows how the story ends, but nobody wants to take their eyes away from this huge sinking boat.

When I see some of the financial engineering currently that were at play, I am a little bothered that if it were a technology used in the real world, it would have been shut down a long time ago. In Nuclear Engineering for instance, we spent so much effort, sophistication and money in having safe and reliable systems that it is mindbuggling to hear about the current domino effect and the inability of the regulators to pinpoints the real problems.

We hear a lot about the fact that CDSs, CDOs and other elements of the alphabetical sauce are contracts that are "difficult" to understand and that this uncertainty is at the center of the mother of all crises. We also hear about banks not trusting each other and how the central banks by lending liquidity are, maybe, making things harder. Yet in the end, there is a sense that some type of worldwide collaboration of different experts is probably the only way to give the beginning of an answer. We'll see on Monday. Irrespective, it looks as though, there is a challenge here that has not been talked about: What type of instruments should regulators have in order to be aware of how good or bad things are ?

Coming back to what put us in this mess, Steve Hsu provides us with a picture of the problem in this entry entitled Notional vs net: Complexity is our enemy, and if you are visual like me you'll understand right away why this is a thorny issue.

In all this, I am more than unsettled with the inability of regulators to clearly identify the areas that need specific control (even during the good times) and I wonder aloud if some of the dimension reduction and CS techniques that we are currently developing and using could not be used in the future to give a sense on bad or good things are to these regulators. Elements that point me in that direction include the following elements:
  • it is a distributed system where some nodes could be required to have counters and other operators.
  • for each of the nodes, the data being processed every day seems enormous
  • there is a need for encryption so that the market and the actors of the market cannot understand the data, yet the final data can only be understood by the regulators.
  • Could we asnwer the following questions: How does disconnecting some of the node change the underlying manifold ? What is a good manifold ?

Any other thoughts ?

No comments: