Risk Management at Lehman

2008-09-15

After today’s turmoil at wall street, shall we calm down to think about the reason?What did Lehman exactly do?Here I quote an article from Martin Hutchinson

Yes, they had risk management, but it mostly used the “Value at Risk” system invented by JPMorgan Chase & Co. (JPM) in the early 1990s. VAR appears to have been designed to let the traders get on with the business of making real money, while at the same time keeping the top brass from worrying too much about the risks traders were taking. It makes a number of mathematical assumptions that are provably false in real life, then assesses the “99% confidence limit” of the loss that may be incurred by each trading position at most 1% of the time (this is usually done by modeling the price behavior of a particular security as a Gaussian normal curve - a “bell curve,” and then taking the point 2.36 standard deviations from the mean).

One big problem with this approach to managing risk is that it doesn’t tell you what can happen the other 1% of the time, when the VAR limit is exceeded. But the top managers of the investment banks were lulled into believing that other 1% didn’t matter: After all, they came to believe, if it was only a 1% probability, how dangerous could it be? However, since VAR was calculated from daily price movements, that 1% actually was quite important.

Even if VAR is modeling monthly movements, an exceptionally conservative use of it recommended by international regulators for theBasel II bank regulating accords, 100 months is still only 8.33 years. Looking at the possibility of going bankrupt every 8.33 years and viewing that as an acceptable risk is absolutely not something that bankers should be doing.

Even if VAR is modeling monthly movements, an exceptionally conservative use of it recommended by international regulators for theBasel II bank regulating accords, 100 months is still only 8.33 years. Running the risk of going bankrupt every 8.33 years is not something bankers should be doing.

The other problem with VAR is that, in most cases, it depends on an assessment of the “volatility” of the security concerned - how much that security bounces around. However, volatility is by definition low in quiet markets and much higher in turbulent markets. Hence, the system’s assessment of risk is low when markets are quiet, allowing traders to pile on positions like madmen, and then zooms upward when things go wrong. At that point positions cannot be unwound.

As you can see, “a culture of risk management at every level” is not a great deal of use if you’re using dozy metrics like VAR to measure risk. The problem is made worse if you indulge in excessive leverage.

Traditionally, the maximum leverage for Wall Street broker-dealerswas held to be 20 to 1. That level was always fudged a bit, partly by pretending that so-called “subordinated” debt was the equivalent of equity, which it obviously isn’t. So, while this 20-to-1 ratio is fine when your assets consist of commercial paper, bonds and shares that can be more-easily valued because they are more-liquid and trade every day, it is far too high when the asset mix has come to include investment real estate, private equity stakes, hedge fund positions, credit default swaps and other derivatives positions that do not even appear on the balance sheet.

Scaling up from 20 to 1 to 30 to 1 - as Lehman did - is asking for trouble.

well, I do not totally agree with this article. I think we should be a little cautious to blame all the issue on VaR. I think there must be some management mistakes or careless. We could always see the moral hazard in using VaR. Usually, when we adjust VaR to 95% confidence interval which means some thing will be happened every 20 days, i.e. once a month, should be very usual. But we also know it is troublesome to explain to the CFO or Cs about the reason every time. VaR won’t tell us how much we are going to expose, but at least it shows the possibility. While, the director of the risk will just simply ignore the VaR fact and change the confidence level to 99%, i.e. once a season. So now everyone is happy?