I read this a while ago … was looking for it in my “notes” (as I refer to the Interesting External Papers category of PrefBlog … and couldn’t find it!
Anyway, Joe Nocera of the New York Times wrote an excellent feature article on Value at Risk: Risk Mismanagement.
The major point to be understood is that management of Goldman Sachs used VaR in an intelligent manner:
in December 2006, Goldman’s various indicators, including VaR and other risk models, began suggesting that something was wrong. Not hugely wrong, mind you, but wrong enough to warrant a closer look.
“We look at the P.& L. of our businesses every day,” said Goldman Sachs’ chief financial officer, David Viniar, when I went to see him recently to hear the story for myself. (P.& L. stands for profit and loss.) “We have lots of models here that are important, but none are more important than the P.& L., and we check every day to make sure our P.& L. is consistent with where our risk models say it should be. In December our mortgage business lost money for 10 days in a row. It wasn’t a lot of money, but by the 10th day we thought that we should sit down and talk about it.”
So Goldman called a meeting of about 15 people, including several risk managers and the senior people on the various trading desks. They examined a thick report that included every trading position the firm held. For the next three hours, they pored over everything. They examined their VaR numbers and their other risk models. They talked about how the mortgage-backed securities market “felt.” “Our guys said that it felt like it was going to get worse before it got better,” Viniar recalled. “So we made a decision: let’s get closer to home.”
Various other elements of VaR and its critiques have been referenced in An Early Debate on Value at Risk.
The main problem as I see it is that VaR does not – and cannot – account for trends. If, for instance, you measure your daily VaR based on data from, say, an environment of steadily increasing real-estate prices, that tells you nothing – NOTHING! – about what happens when they decline. Especially if they decline suddenly and interact with factors not in your model, such as “jingle mail”.
And the other problem is – as Taleb appears to have made a career out of saying – fat tails and black swans.