Cheng, I fully agree with you and argued softly on this line during the talk...
Basically I was there the only "disrespectful" listener all the audience (purely academic and most of them fully unaware of credit risk as a topic) was so captivated because you know Duffie is "Duffie" ...
however as I said in the previous post he alluded to the fact that
"they (?) sort of (?) could be better at pricing what (?) at ML because of that model whereas.. all the others (?who) not yet enlightened (...) were using copulas..."
that he was increasing accuracy... that could be useful even if it is a "mini part of the unknown"?
indeed he is good at marketing
И ветер, и дождик, и мгла Над холодной пустыней воды.
"they (?) sort of (?) could be better at pricing what (?) at ML because of that model whereas.. all the others (?who) not yet enlightened (...) were using copulas..."
This may be too "point to the scoreboard" but ML it is looking like they are writing down $8Bn more bringing their total to $30Bn so whatever they've been doing clearly isn't working better.
You can throw away all your he-man theories. Once, you've lost that grubby feeling.
I really don't get a lot of pleasure from their foibles or those of the street in general. It makes things bad all around. You are right on the account that he is a good marketer. Only a good marketer could with a straight face, assert he has edge while his institution is busily writing down assets in a sector where his edge is supposed to apply.
You can throw away all your he-man theories. Once, you've lost that grubby feeling.
researchers have personality too, especially those from biz school, they need to be able to sell their ideas. On the contrary, I have seen quite a few brilliant yet humble fellows in math.
something I am really mulling in my mind right now is the idea of bankruptcy as a means to stop the race to collateral and introduce some civility to priority rights, as compared with credit trading in general, before the event. A lot of credit trading is sort of like front-running events, changing your position before everybody does and the common information puts everybody on the same footing... i.e. doing what would happen if BK were not around. This is similar to the situation where latent variables might appear - i.e. there are different strategies about positioning in order to front-run the event (i.e. 'GS hedged their subprime' vs everybody else who followed things right to the end and will get the 'post-bk' collective result).
I've made the whole thing very vague but I what I'd like is to get a better mathematical grip on how this idea impacts the underlying timeseries. I.e. everybody knows single-name credit is gappy in an odd way. Lev supersenior was even gappier b/c the market is determined by far fewer players. And, I think for ABS the jury is still out in terms of AAAs and whether it is better to stay long and try to repo the stuff to the relevant central bank, or to cut losses now and not accept the collective result.
Finally, I've said it a million times but I do think Soros' alchemy point about when the analysis of any situation gets reduced to calculating certain benchmarks, the emphasis will shift to delivering the benchmarks and nothing else. I think ABS CDO was a particularly bad offender - it was pretty obvious that this thing levered 100x had a serious ALM mismatch but it got AAA anyhow. It may be the case that everybody knows the situation is wrong but they haven't put their order to front-run everybody else yet.
to be honest a little ... it needs de-KR-ypting.. you loose me totally on the first sentence of the first paragraph and a little at the end ... but still some of your idea gets through..
to get a better mathematical grip on how this idea impacts the underlying timeseries
that would be cool indeed.
И ветер, и дождик, и мгла Над холодной пустыней воды.
Aaron, can you indulge us with your response to Taleb's assertion/critique that your risk models are worth 2c on the dollar? (referencing the "debate" that ensued following Taleb's lecture according to the new Bloomberg magazine article ). I've just always been curious as to what a professional would respond with when faced with the "gaussian statistics and risk models bad". Well it wouldn't be a random event if we could capture it now would it?
Gaussian is not useful for risk management. Thin tails and only pairwise dependence make it inapplicable.
The key to a number is not how you compute it, but how you test it. There are two kinds of useful risk management numbers. The first is validated by a hedge. If I draw a graph of my predicted price tomorrow versus certain market variables, you can check tomorrow's predicted price against actual. If a model works reasonably well for an extended period of time, you can put some reliance on it. It is, of course, possible that it will stop working suddenly. So it only tells you about the center of the P&L distribution, not the tails.
The other useful number is validated by a non-parametric backtest. If I give you a 95% one-day VaR every morning for 500 days, you can check to see if there were 25 (plus or minus 10) days had larger losses than the predicted VaR. You can also check to see if the VaR breaks seem to be distributed independently in time, and independently of the level of VaR. If it passes the tests, you can put some reliance on it. Again, it only tells you about the center of the distribution.
Risk measures that are not backed by one of these two methods are useless.
Within the center of the distribution, you have lots of good observations and you can use all kinds of statistical methods with confidence. Since we spend most of our time in the center, that's important.
Taleb's valid gripe is that some people look at the risk management definition of the center treat it as the worst case. It's only marginally better to assume the worst case is a bit beyond the boundary, say three times VaR. It's essential to realize that outside the center, all bets are off. Your losses can be 10 or 1,000 or a million times VaR, or more. The tail includes things like market shutdowns (VaR explicitly assumes normal markets), faulty position information (including rogue trading and fraud), expropriations and other things. You can't analyze the tail using the techniques that work in the center. And you can never check your predictions against the past, there is not enough data (if there were, we wouldn't call it the tail).
Outside the tail you get three kinds of events. The most common is unlikely combinations of everyday events, sometimes you get 65 heads out of 100 independent, fair coin flips. As a rule of thumb, these tend to be within 30% of the limit you established for the center and occur between once a month and once a year depending on your application. Next most common are periodic events like credit crunches, stock market crashes and so on. These will affect a typical application a couple of times per decade. Both kinds of events can be analyzed using long-term qualitative data from a wide variety of markets; within the center of the distribution we rely mainly on short-term quantitative information from the specific market.
Finally, rarest of all, are Black Swans. These also seem to come along a couple times per decade or so, but they affect all markets at once, so there are fewer of them than the better-defined disasters like credit crunches. I think the current liquidity crisis qualifies, although its nature has yet to be defined. Before that I'd put 9/11, LTCM (which is really shorthand for a bunch of stuff) and 1987 stock market crash (again, the event was much bigger than that description). Actually, in some ways it's tempting to combine the middle two and say there were deep and unexpected financial changes from 1997 to 2002.
Taleb says, and I agree, that it's pointless to try to predict the next one. However you can do things to be well-prepared in general. You can do sensible things to help survive the next Black Swan. Hiring smart people, having good morale, sensible IT and HR policies, honesty, practicing emergency measures and other things might help. If the earth is wiped out by a meteorite, then none will help. If the next Black Swan is a computer virus that disables the world's computers overnight then some will help and some won't. But they're still worth doing.
Aaron: I think those were the best eight paragraphs on the usefulness and limits of risk measure and management ever written, and I hope you won't mind if I share those comments (with attribution of course to your good self) with some folks that need to read them.
I always find horizons and transition spaces fascinating, and so your comment "It's essential to realize that outside the center, all bets are off" appears to me to have another dimension of inquiry (not that we want add dimensions and gain a curse ). That debate to me is: we know what the "center" is, we even have a precise measurement for it, and we can say what the boundaries are, but do we dramatically go from a roughly tractable world (Gaussian soft rationalists) to a completely intractable world (Taleb's argument) once we cross that horizon? Are all bets really off?
I hasten to add that you folks who are much much better in math than me, and know about topology and limits and bounderies and creases from every math discipline may find this naive. But it is hard for me to believe we really live in a world where "we know the center well, we are comfortable in the center, we can measure the center, but once we are out of the center we are in "The Land of The Lost".......it just strikes me as too pre-Columbian and flat earth.
Okay, if I can turn a sphere inside out with smooth isotopy, how come I can't turn the manifold that is myself inside out to see why my stomach hurts?
and argued softly on this line during the talk... 
). That debate to me is: we know what the "center" is, we even have a precise measurement for it, and we can say what the boundaries are, but do we dramatically go from a roughly tractable world (Gaussian soft rationalists) to a completely intractable world (Taleb's argument) once we cross that horizon? Are all bets really off?