Each of these market failures has been exposed by events over the past 18 months. When risks materialised outside of calibrated distributions, risk models provided little guidance in identifying, pricing and hence managing them. This failure is not of purely academic interest. The breakdown of risk models is itself likely to have contributed importantly to crisis dynamics. Why?
First, the potential losses arising from under-pricing of risk are large. …
Second, the breakdown of these models had the consequence of turning risk into uncertainty, in the Knightian sense.11 Once the models broke down, how were assets to be priced? Practitioners have a devil of a job pricing assets in the face of such uncertainty. So too do academics, though some attempts have been made.12 The theory of asset pricing under Knightian uncertainty throws up at least two striking results. First, in the face of such uncertainty, asset prices are not precisely determined but instead lie in a range. This indeterminacy in prices is larger the greater is uncertainty and the greater agents’ aversion to it. Second, asset prices exhibit a downward bias relative to fundamentals. Uncertainty gives the appearance of “pessimistic” expectations.
Apparently the reason for this abrupt outbreak of Knightian uncertainty is that bankers have suddenly realized that there is a difference between reality and their models. As long as the world behaves according to model, bankers want to claim that they are earning profits from managing “risk,” and as soon as their models fail, risk becomes uncertainty and necessitates a government bailout.
In short whenever you read that bankers can manage risk, but uncertainty requires government intervention, you should hear: “Privatize the profits and socialize the losses.”
The truth is that bankers always have to price assets in the face of Knightian uncertainty, they have just chosen to spend the last decade pretending that this was not their job.
Note: The difference between Haldane's approach and mine is that Haldane believes that is possible to change the models so that they will be "roughly right," whereas I believe that the fundamental flaw was the belief that any model could be "roughly right." Don't get me wrong. I'm not a Luddite. I don't think we should stop analyzing data using models. I just think that every time you use a model, you need to be aware of the many ways in which it is totally and completely wrong. In short, models are never intrinsically valuable, but careful and constructive human interaction with models can be priceless. Update 2-25-09: Have finally read Haldane with some care. This may not be so much of a difference. Haldane's just talking like a regulator who has to design concrete tests for banks to perform. He might agree that the ideal is to keep updating those tests regularly as regulators' understanding and experience changes (and to keep the banks on their toes).
Update 2-15-09: There is a problem with Haldane's use of the term Knightian uncertainty (and probably with the bankers' use of it). Haldane's paper argues that bankers measured uncertainty incorrectly. He then states "the breakdown of these models had the consequence of turning risk into uncertainty, in the Knightian sense."
This statement confuses Knightian or unmeasurable uncertainty with unmeasured uncertainty. The bankers' models were wrong, they mismeasured risk. The bankers apparently wish to claim that this unmeasured uncertainty was unmeasurable -- that is why they call it Knightian. Haldane by contrast argues that this unmeasured uncertainty is in fact measurable by using a broader range of data and stress testing.
In short, when bankers claim that Knightian uncertainty has increased, they are just trying to make excuses for their own failures. Haldane has recognized the banker's failure to properly model risk, and doesn't appear to be using the term "Knightian uncertainty" in a manner consistent with its original meaning.
No comments:
Post a Comment