Will Ferrel's new movie, Land of the Lost, is a comedy in which a mad scientist is catapulted back in time, as a result of a glitch in his time machine (surely an operational risk event). Marooned in pre-history, Ferrel is manhandled by Mastodons and pterrorized by Pterodactyls, but in a funny way.
Talking of dinosaurs, the Basel Committee has just released a new report on progress towards Basel II.
This report discloses results from the 2008 Loss Data Collection Exercise (LDCE) in which banks around the world were asked to supply details of the losses they experienced due to operational risk and to provide information on the methods they are using to calculate Operational Risk Capital (ORC) under Basel II rules.
The new report was published by the 'Operational Risk Subgroup of the Standards Implementation Group' or SIGOR. This is a shame because SIGOR has taken over from AIGOR, one of the great regulatory mnemonics, conjuring up pictures of mad Transylvanian butlers in Mel Brooks comedies. AIGOR unfortunately produced only one report before expiring - RIP AIGOR. On the other hand, SIGOR sounds like a Norse god, maybe based on the Roman Janus, the two-faced god who doesn't know whether he is coming or going?
[By the way, if the Basel Committee is looking at implementing risk management 'standards', why don't they just mandate ISO 31000 and be done with it?]
The LDCE surveyed 119 banks around the world on their operational risk practices but the analysis concentrates mainly on the practices of the most sophisticated banks, i.e. those that are AMA (Advanced Measurement Approach) complaint [or near there given that the US is not on board officially with Basel II yet]. In all, 42 AMA banks responded to the survey request; this is a number that is on the borderline for statistical analysis of 'small samples' but does just squeak in under one of the golden rules for surveys that the number of survey responses should be greater that than the number of people doing the surveying. SIGOR has 40 members so there is just over one AMA bank per committee member.
The results of the 2008 LDCE survey are 'ho-hum', less 'Planet of the Apes', more 'Much Ado about Nothing'.
Whereas Credit Risk is the flesh-eating Raptor in the Jurassic Park of risk management, Operational Risk is the small shrew-like fossil that is crushed for eternity under the back foot of a much larger Brontosaurus. Rat-like, Operational Risk is small but always there, annoyingly hard to get rid of, but cannot be ignored because it may gnaw away at the firm's foundations (Barings, AIB etc. etc.)
The 2008 survey confirmed (for the umpteenth time) that operational risk losses consist of a very large number of small losses and a small number of very large losses. Unfortunately, in practice there are so few 'real' large losses for meaningful statistical analysis that they have to be invented - correction - have to be estimated using Scenario Analysis. To quote the report: "At many banks, the number of scenarios greater than Euro 10 million is approximately twenty times larger than the number of internal losses that are greater than this amount." 20 times - enough said - it is Fantasy Island!
The survey found that AMA banks report that around 11% of their total regulatory capital is allocated to Operational Risk, which is a long way from the 20-25% originally estimated when Basel II was first envisaged. [Interestingly, the percentage provided in the report is a surprisingly precise 10.8%, surprising given the gross numbers provided by the banks surveyed. This false precision is enhanced by the use of 10 (yes TEN) digit currency exchange rates for converting loss data into Euros. These rates were selected on the somewhat arbitrary date of 31st March 2008, so results do not reflect any currency volatility over the 3-5 years covering the operational loss data collected].
The report gloats that the 11% ORC capital estimated for AMA banks is (much?) less than the 15%-18% capital charged to non-AMA banks that apply the mandated Basic Indicator and the Standardized Approaches (i.e. BIA & TSA banks). If I was a non-AMA bank I would be extremely annoyed at this discrepancy and would demand that Basel immediately reduces the ORC 'tax' imposed on non-AMA banks, given that the report found, not unexpectedly, that AMA banks have considerably more operational risk than their non-AMA counterparts. Less Risk - more capital; not a good look?
Having published 50 pages of detailed tables and 97 pages of commentary, the report concludes that not much has changed since the 2006 LDCE survey; and nobody, but nobody, has found the secret of quantifying Operational Risk Capital.
For the geeks amongst us, the results are interesting (if only mildly).
The report found that banks overwhelmingly (93%) apply the Poisson distribution to model the 'frequency' of operational losses. This distribution is employed mainly because it is easy to use, but some studies* have concluded that operational data does NOT follow a Poisson distribution! Given that SIGOR has access to the cleanest, most comprehensive set of Operational Risk data ever collected it should be relatively simple to test the hypothesis - Poisson or not? However, SIGOR did not choose to do that straightforward test, wimping out with a note that this distribution 'is broadly aligned with supervisory expectations', whatever that means. The Poisson is a fishy fossil like the Coelacanth, the so-called Dinosaur Fish, which survived undetected for eons in deep water; Poisson just goes on and on, and on, and on.
Similar wimping out occurred when modeling the 'severity' of operational risk events. While 31% of banks use Extreme Value Theory (EVT) to model the tails of severity distributions, 69% don't, preferring other models or defaulting to the catchall Lognormal distribution. Again with a mass of data, SIGOR could have shed some more light on which statistical distributions may be defensible in the tails of operational risk distributions but they chose not to do so.
On the subject of 'dependency', however, the geeks go to town. Notwithstanding the fact that it would be difficult to justify any correlation between, say for example, operational risk in Corporate Finance and Retail Brokerage business units, copulas are 'soup de jour'. SIGOR found that 43% of 'advanced' banks employ copulas to model dependency between operational risk types, predominantly using the 'Gaussian Copula'. Of course, the survey was taken in early 2008, well before the publication of the now-famous article 'Recipe for Disaster - The Formula that Ate Wall Street' about the problems experienced using the Gaussian copula in the much more tractable area of Credit Risk. Writing after the article in 2009 however, SIGOR points out, somewhat tongue in cheek, that the Gaussian copula "may underestimate the probability of joint extreme events", chuckle, chuckle!
There are further hints of humour in the report. It is reported that 2% of AMA banks (in reality one joker), employ a 'Zero Copula' to measure operational risk dependencies. Now a Zero Copula is a phenomenon in linguistics (!) in which a verb (usually a tense of 'to be') is deliberately omitted from a sentence for effect, as for example in "Basel II, too little, too late". So either copulas are breeding like Lagomorphs (pre-historic rabbits) or SIGOR has been punk'ed. Either way, as Fred Flintstone would say 'Yaba Daba Doo-Doo'.
Overall, the observations in the report on 'observed practices' are of the form: this is what happens, we are not happy, but we don't know how to do it better, let's move on; next agenda item, which international resort should hold the next SIGOR meeting?
In the pre-history of regulation (i.e. before the Global Financial Crisis is about the change the rules) Basel is the vegetarian Diplodocus, huge body, tiny brain, and easy prey to rapacious Tyrannosaurus Rex such as Goldman Sachs and JP Morgan. In years to come, students of finance will look at the fossilized Diplodocus (Basel II) and marvel that it ever managed to grow to adulthood, never mind dominate its environment for such a relatively long time.
* A good analysis of the problems surrounding operational risk data can be found in "Scaling Models for the Severity and Frequency of External Operational Loss Data" by Dionne and Dahen (HEC Montreal 2007)
