Monday, May 14, 2012

The Role of Risk Models in Decision Making about Loss

In a Financial Times opinion, Frank Partnoy, law professor and financial market regulatory expert, identifies computer models as a main culprit in creating false perceptions of risk in the recent $2 billion loss by JPMorgan Chase.  He says that the models were also partly responsible in several other historic losses: Barings Bank (1995), Long-Term Capital Management (1998), Enron (2000), and Lehman Brothers and AIG (2008).  He argues that the single modeled value-at-risk number is not sufficient to describe the risk to investors.  

In each case the victim was thought to be a well-run, relatively safe institution. And in each case the culprit was financial innovation and mathematical models that wrongly suggested positions were low-risk. Each of these firms traded complex financial contracts and then used a single number – “value at risk” – to estimate its maximum probable loss, just as a weather forecaster might estimate the likely cost of hurricanes during an upcoming season. 
Partnoy argues that the problem with the current system is that regulations have failed to push financial institutions to reveal additional information so as to better disclose the full risk.
The law should require JPMorgan to tell investors what would cause a $2bn loss. 
law should punish anyone who defrauds investors by citing one value-at-risk number and then losing 30 times that amount.
Such solutions make some assumptions.  Most notable, it assumes that producing more information would result in better decision making about the risk.

Blogging on Finance at Reuters, journalist Felix Salmon, discusses much of the same, but suggests that the problem is not a lack of information but the way the information produced by the models is used in decision making about risk.  He argues that by this point in time, the models are well known to be "faulty" but
The problem is that pretty much by definition, it’s impossible to model model risk.
Depending on assumption inputs, the models can produce any number of loss probabilities and loss events.  This is seen clearly in comparing catastrophe models.  It is ultimately the human user that makes judgements about which model to use and what the modeled risk means for the decision at hand.  Salmon suggests that it is the responsibility of upper management at financial firms to consider the risk of the model to the larger system it is being used to help manage
I know that your highfalutin’ models say that these exposures are risk free, but I don’t understand how this isn’t risky, so go unwind this trade 
Salmon resolves a set of pragmatic "dumb rules"
Your sophisticated platform needs to be built on a foundation of dumb rules: simple limits on how big any one position can get, on how much exposure you can have to any one counterparty, or in general on any trade which is based on the hypothesis that your desk is smarter than anybody else on Wall Street.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.