CDO waterfall errors

CDOs are complicated. In particular, many of them have waterfall structures that include diversion tests - if x then tranche y gets some money, otherwise it goes to tranche z. Unsurprisingly, trustees sometimes get these tests wrong. Expected loss has found an example: I do encourage you to read it if you have an interest in either structured finance or operational risk.

Fear vs. Operational Risk

The FT reports that France's Finance Minister Christine Lagarde wants bigger penalties for banks who are found to have weak controls in the wake of the l'affair Soc Gen:

France’s banking regulator must be allowed to impose far higher fines on banks that fail to monitor carefully the risks taken by their market traders, according to a government report on the rogue trading scandal that has rocked Société Générale.

I can see that there is a profound sense that something must be done, but I doubt raising fine levels will do much good. If the risk of losing a few billion euros is not enough to persuade the banks to invest in solid controls then it is hard to see that fines will do much good. The reality is that we have not had a big rogue trader event in a bank since Leeson and Iguchi in 1995 - John Rusnak does not really count - so some banks have become complacent. The most effective pressure would be disclosure, but I rather fear that the only result of having to disclose supervisor's quality of controls scores would be to dilute their content.

As a side note, I wonder what effect the Soc Gen loss is having on industry operational risk loss databases. One would hope that it would be widening tail estimates and so capital would be going up.

What does safe mean?

It is an interesting question. Nothing is safe, 100% robust under any set of circumstances. If a two hundred foot high sea monster climbs out of the Thames and starts munching on Canary Wharf a few disaster recovery plans would doubtless be found wanting.

There are at least two issues. The first is to encourage people to be skeptical about the performance of any construction, mechanical, electronic or intellectual: there are some events that will screw up any design.

But then we come to the problem of how to estimate how unlikely these testing circumstances are. Typical operational risk events involve a concatenation of errors, of individually improbable circumstances. Sadly it seems that sometimes these events are not independent so that the joint probability of a screw up is much bigger than one might think. For that matter, the equity, credit, FX and interest rate markets often have low return correlations: but they can all move together in a crisis, as LTCM found out. It isn't that a plausible worst case is bad -- we knew that -- it is that the worst case can be much more likely than it appears.