Good Reporting

Here's a link to an excellent article in the Globe and Mail which provides some insight into the interaction between organizational culture, mathematical modelling and risk taking at financial institutions. In case you are interested in that kind of thing.

A few quotes since it will be paywalled soon enough:

"In all financial institutions, there is a daily battle between the risk takers and the risk managers. The takers push for bigger positions to make bigger profits, while the managers push for prudence and caution.

But as their warnings of potential loss were proved false each day by the soaring financial markets, many risk managers lost the ear of management teams focused on the vast profits generated by the people in the business of creating the structures. That led banks to take bigger and bigger bets."

...

"But some of the fault also lies with risk managers who relied too much on their tools, the statistical models, which were rapidly eclipsed by the rapid innovation in financial markets that begat complicated structures such as CDOs, so-called CDO squareds and structured investment vehicles (SIVs).

“Risk management tools are blunt instruments, which calls for prudence,” said Louis Gagnon, a former Royal Bank of Canada risk-management executive who now teaches business at Queen's University. “If you know you are driving your car on a foggy evening, you are supposed to go easy on the gas, but it's not necessarily what happens.”"

...

"Along with correlation, another term has come to haunt risk managers: “tail risk.”

It's an odd name for the statistical chance that returns on any given investment will fall outside the normal probability of events. (When plotted on a graph, the statistically probable events are grouped in a bell curve, but there's a long tail of improbable events that trails off to one side, hence the name.)

In other words, most of the times markets behave normally. But every so often they don't. Those abnormal events fall in the “tail” of the risk curve.

Many risk managers, especially those at banks, use the normal probability concept to develop a yardstick called Value-at-Risk (VaR), which measures the amount a position taken by traders could lose on any statistically “normal” day. Normal is defined as a move of less than three standard deviations from the mean, and the assumption is that normalcy will reign for all but one day in a hundred, or even a thousand. That's when the tail comes into play.

Most banks look back three or four years to determine the likelihood of loss – meaning that just before last summer's blowup they were looking only at years of unnatural calm. Markets fooled the models.

“The tail events happen far more often than we would predict,” Mr. Gagnon said. “But what are the predictions based upon? The normal distribution of events.”

As a result, VaR failed investors. For example, CIBC had a daily VaR in the third quarter of 2007 that averaged $9.9-million, according to the bank's quarterly investor presentations. Yet three times in that quarter, as the credit crunch picked up steam and the bank booked writedowns, it lost more than that in a single day, including one loss of $120-million."