Tuesday, February 24, 2009

Financial models, variables, parameters, assumptions, "Garbage In, Garbage Out"

It is so easy to blame so much of the current financial crisis on complex "financial weapons of mass destruction", but it is not necessarily so simple. I have read descriptions about how CDOs (Collateralized Debt Obligations) are supposed to work. The actual ideas are not that bad. Clearly the models for these instruments ultimately failed, but as far as I can tell the basic models are in fact reasonable.

Like all models, there are plenty of variables and parameters. If you do not get the variables and parameters right, then of course the models will fail.

Any wizened old computer programmer can tell you a simple truth about even the most perfect computer software: GIGO - Garbage In, Garbage Out. In fact, it does not matter if you get 99.9% of the variables and parameters absolutely right, even one bad key parameter can spoil the entire model.

In the case of CDOs, a key parameter is the maximum default rate, how many people can fail to pay their mortgage payments before the model begins to fail. If the default rate is below some threshold, the models work extremely well and almost everybody makes lots of money. But, if the default rate is above some threshold, the models, as they say, "blow up" and either nobody makes any money (a CDO of absolute junk subprimes) or many people get wiped out and some indeterminate number are still magically protected by the models. But unless you can accurately predict how many will be protected, the safe assumption is that none will be protected. Unfortunately, we need to know the expected default rate in the future for a given pool of securities, and that is the one most important thing we do not know at this stage, unless the government steps in with a guarantee.

The crisis appears to have occurred because the assumptions about the near future and resulting parameter values were simply wrong, really wrong. Sure, the modelers may claim that they did not "know" that junk mortgages were being sold and that the housing boom was actually a bubble, but models are supposed to incorporate all of the variables and parameters of the real world, not some idealized world.

During the housing boom, modelers set their parameters on the assumption that the future would resemble the past. No gung-ho modeling group anxious to share in the corporate bonus pool was going to set their parameters based on the near future resembling The Great Depression (or worse.)

Wouldn't it be a hoot if it was all that simple, great models, but just a few "bad" parameter values?!?!

-- Jack Krupansky


Post a Comment

Subscribe to Post Comments [Atom]

<< Home