This page has been archived and commenting is disabled.
Why VaR Is A Joke: Morgan Stanley Admits Losses in April And May Were "Much Higher" Than Anticipated
Zero Hedge has long contended that risk models based on VaR "predictions" are flawed and only add to systemic instability due to the ever increasing correlations across all asset classes. We now read a first hand mea culpa from Morgan Stanley's Jim Caron, in which the head of the firm's rates strategy highlights precisely this problem: the complete collapse of predictive models when multiple sigma events like the May Flash Crash and the accelerating sovereign collapse of the past several months occurs: "April and May were difficult months for us and others, judging by fund data on market performance. We did not properly discount the risks associated with peripheral Europe. As a result, we had a larger risk exposure than we should have. We measure the return potential for our positions on a per-unit-of-risk basis, similar to a Sharpe Ratio. That unit of risk turned out to be much higher than we anticipated. This will force us, and many others, to right-size our risks." We wish we could agree with the last statement. Alas, each and every risk management group at comparable prop trading desks (to that of Morgan Stanley), will undoubtedly chalk off recent events to chance, and as these "will never recur", business we will promptly return back to normal, until we see another record crash in the Dow, only this time not 1,000 but multiples thereof.
As for those few remaining sane individuals who do learn from past mistakes, here is what Jim Caron suggests are the key implications of consistent underappreciation of risk:
Liquidation of risk exposure: Portfolio positions turned out to be much more correlated than we had initially anticipated. Traders seek to reduce correlation by liquidating many positions, leaving behind perhaps only a few core positions. We saw these liquidations in May and early June
Sit, wait and re-assess: Traders will now have to evaluate the new risk relationships. Since there is great uncertainty, traders might start by making small and short-term tactical bets to get a feel for the risks. Again, only a few core positions may still be left on.
Right-size risk: As the new market environment becomes somewhat better understood – albeit still marked by great uncertainty and higher realized volatility – traders could now start to make an assessment on the proper risk they should have relative to the increased level of expected volatility of returns. For example, if the market is twice as volatile today as it was before, then one should run positions with half the size of risk.
For better or for worse – the introduction of a new tail risk: Given the losses taken and positions liquidated in the past few months, the new tail risk is for risky assets to reverse sharply higher and for yields to rise. This could cause traders to chase performance, so as not to be left behind relative to their peers. Similarly, if markets turn against them, then they will be quick to exit. This introduces two-way risk: traders may start to react equally to both good and bad news. Previously, the tail risk traders were mostly focused on a worsening of risky assets. Now they have to be concerned about both tails, for better or for worse, which will add to market volatility.
Incidentally we agree with virtually all of Caron's observations. And we suggest that before the travesty of a FinReg piece of toilet paper is signed by the president, that someone with half a brain at least pretend to recommend some provisions to haircuts in bank risk exposure. On the other hand, if indeed the Volcker Rule in its full form is set to pass, and the only risk banks are taking on is that compensated by a commission, once prop trading groups are finally split off, a development we have been urging since May of last year, all this is moot, as no longer will banks have to worry about old tail risk, new tail risk, or any risk: they will be merely hedged books, operating in a liquid (thank you HFTs now in equity and soon in cash and CDS) market, where 6, or 666 sigma events will have no bearing on the banks' requirement to be bailed out be taxpayers imminently.
»
- 8461 reads
- Printer-friendly version
- Send to friend
- advertisements -
I find it interesting that someone without a background in the subject can see all the faults of their logic. Duh, you leverage up something and the tail risk goes up as the percent of leverage. The asses (Bernanke in particular) has no clue how markets actually function. I see him as kind of a Mr. Mago figure. the bankers cry that the market is acting in an irrational manner, yet when you lever up the smart thing is to get out of the position as soon as you can. Yet they do all they can to encourage leverage. How f"ing" stupid can you be.
I hope some aliens from outer space see our misery and remove them as a favor to man kind. please dump them on some out of the way asteroid with no way back!!!
They learn as they go. I bet you that for the next crisis, they'll do a bit better :)
You gotta love working around jokers like this guy...
It must be comforting medicine to believe in easter bunnies, tooth fairies, and economies that conform to theories concocted in some laboratory.
Just get out the mass spectrometer and measure the asset spectral lines and then 'right size' the risk...lovely.
After the Howie Hubler mess, they still can't calculate VAR.
." We wish we could agree with the last statement. Alas, each and every risk management group at comparable prop trading desks (to that of Morgan Stanley), will undoubtedly chalk off recent events to chance, and as these "will never recur", business we will promptly return back to normal, until we see another record crash in the Dow, only this time not 1,000 but multiples thereof.
___________
As a result of the flash crash, many market participants have Hail Mary limit orders set to take advantage of the next one.
That's why there probably won't be a next one for a looooong time.
I'm not so sure with this volume.
They see the orders...they can position themselves accordingly to take advantage or let the orders expire...if they want "other peoples money" they will have to do it on other peoples terms ;-)
What do you want to bet the koolaid drinkers have their 'hail mary orders' set just above the average percentage drop where the trades got cancelled? When the real flush happens, they will get filled before they realize the circuit breakers are now cancelling their 'hail mary' sell stops? Talk about your six sigma pig with lipstick....
What was that about the Bank had to changes its VaR model every quarter when so many days of losses exceeded its one day Var expectation. Was all that in the last century? What do they today? Mark to guess and by golly?
Somehow these quackrobats have interpreted black swan to mean inexplicable tsunami. The point is what some may think is impossible is indeed very possible. Mr Taleb just loves VAR.
The metric for mayhem.
Such a delicious term VaR.
Damn near all V is at R here.
Hair splitting financial mumbo is just going to get stomped by the Big DD Jumbo.
I say you take that to the market and trade it.
We are now firmly in the grip of the Punditocracy and they are all nekkid.
Dog save us all.
ORI
http://aadivaahan.wordpress.com
Thre's no excuse for this - the amount of work that has been published on expected drawdowns, predictability of crashes, correlation going to 1 in a crash, volatility behaviour, skew behaviour, equity-credit correlations, equity vol-credit correlations... MS was supposed to be the conservative, client-friendly one these days. Maybe they just aplied that strategy because they realised they were incompetent with risk.
I wasted 5 years of my brain as a quant. I wasn't very smart, it took me that long to realise it's all bullshit. The job is just to get the trader/manager the number he wants, any way you can. With pnl, he wants a positive number. No matter what happens in Greece. With risks, the number he wants is the same as yesterdays. No matter what happens in Greece.
It makes me sad.
A big Taleb fan but not exactly a higher math guy, I recently tried to get my arms around VaR, which caused me to re-examine the nuts and bolts of statistics.
So you have a set of values (data) which has an average value or mean. The variance is the average of the squares of the distances of each value from the mean. Standard deviation is the square root of the variance.
Ever wonder why all those distances from the mean need to be squared in the first place? Well one reason is to get rid of any negative values which keeps the subsequent math simpler. But you could do that by just noting the absolute value of the distance from the mean (if the mean is 2 then -3 and 7 would both have a distance of 5 from the mean).
Doesn't all that squaring and square root finding mess things up in other ways? Why not just have an absolute value derived mean deviation?
I found a paper by a Stephen Gorard from 2004 (published in between Taleb's two popular books) which addresses this issue: http://www.leeds.ac.uk/educol/documents/00003759.htm
It turns out that SD in a "normal" Gaussian (bell-shaped curve) distribution in sample is a more stable indicator (than mean deviation is) of it's equivalent in a larger, general population.
But in longer tailed distributions, outlier values become very large when squared (which square-rooting doesn't completely mitigate). Add to that the modern tendency to eliminate outliers in the first place and SD becomes not so great when used to measure the risks of rare, outlier events.
This seems kind of obvious now but it's never really mentioned.
Standard deviation, sigma and it's derivatives get thrown around so much but they just don't seem to be that appropriate outside of well-behaved data sets. If anyone can point to other such research along these lines I'd much appreciate it.
Why not just have an absolute value derived mean deviation?
Mathematically, you're asking, why not use an L1 norm?
Not being a quant (engineering is my thing), I assume that the L2 norm (your standard deviation) is more convenient theoretically.
I don't think it's really more convenient but it makes things neater when dealing with normal distributions. In longer tail distributions you get into trouble when you try to tidy things up like that I think.
Didn't LTCM purport that they were hedged to 9 sigmas or some such craziness?
Then Russia defaulted. I mean if that isn't a 10th sigma event, what is? Because countries NEVER default, right? It's absolutely of almost negligible statistical probability that a country would default.
The tails are significantly larger than any of these idiots want to admit. Certainly, backtesting to the start of the 82 rally suggested that there WERE no datapoints, but rearward bias is NOT predictive statistics!
Yes about the '9 sigma hedge' (though i dont recall if 9 was the official # given) but the problem with LTCM was they kept piling on their convergence plays even after the market was in or around the 99.5% mark prob of 'this shit never happens'. Averaging down, if you will. Even though the trade still kept going against them.
Without rules a manufactured system seeks entropy. The markets have demonstrated that there are no rules. Maximize entropy and you get chaos. Have a nice day!
These financial geniuses need some common sense.
As suggested earlier, the EURUSD daily chart is giving bullish signals.
http://stockmarket618.wordpress.com
http://www.zerohedge.com/forum/latest-market-outlook-1
Tyler, how the fuck is this braindead turd still allowed to post in the comments? DELUSER.
After giving it much thought (not), I've adblocked your idiotic flashing gif avatar.
Too much math , not enough brain cells