The Scam Of VaR "Risk Models" in [Market-Ticker]

Posted 2012-05-18 08:28
by Karl Denninger
in Editorial

The Scam Of VaR "Risk Models"

The value at risk for equity exposure discussed above is based on J.P. Morgan's RiskMetrics(TM) approach. Both value at risk methods utilize a one-day holding period and a 95% confidence level. Cross-commodity correlations are used as appropriate.

Read that.

It goes back to Enron, but it points out the problem, if you think about it. Take a few seconds before you continue reading and see if you can figure it out.

Got it?

It's right here: "Both value at risk methods utilize a one-day holding period and a 95% confidence level."

What if you cannot exit your positions within one day, and there is a 5% chance that the model fails on any given day, or one day in 20 anyway?

In either case the entire exercise is a scam; neither of the boundary conditions is reasonable. Not only is a 95% confidence interval ridiculously wide (this means that on average one business day a month will go wide of the mark!) predicating the performance on a one-day holding period when one cannot unwind the positions involved within a period of weeks or even months without adversely moving the market against you is knowingly using a bogus metric to produce your claimed "value" model.

But it gets worse: Statistical models are all about random distribution of events. The problem with such models is that markets are, in the "tails", not random at all. The premise that a large enough group of people produce statistically-randomized outcomes is only true until "something happens" and the actions of participants become cohesive (e.g. a bank run) -- then your entire model goes out the window as the very predicate on which it was based is invalid under those conditions.

This is what we live with today because we allow institutions to set "risk models" and then effectively counterfeit credit predicated on the "reasonableness" that it will all turn out ok instead of forcing them to hold actual capital against each and every position.

The consequence of that counterfeiting was this:

The credit created continued to expand for 30 years on the back of this bogus model, inflating asset prices to ridiculous heights. That game continued right up until someone couldn't pay, at which point the fallacy of these statistical models came apart in the face of the people who had been using them and the rubes in government and the public who had been comforted by them.

But rather than learn our lesson and force those who did the dumb things to face their follies and eat them we twice "doubled down" with more idiocy and at the urging of people like Bernanke who has admitted to and is conspiring with the government in "supporting" these bogus valuation models. The Internet Bubble was the first iteration and the Housing Bubble was the second. Now we're trying for a third but we've already transferred the risk to governments -- and as we are learning the hard way when that gambit fails, and on these statistical models it absolutely will, there's nowhere else to turn and "fun" situations like Greece and the entire EU entitlement and banking systems go into the meat grinder.

We're right behind them folks, and our time to put a stop to this crap is rapidly running out.

Discussion below (registration required to post)

Main Navigation

Full-Text Search & Archives

Archive Access

Blogtalk 3:30 CT Mondays

Items To Look At

Discuss The Capital Markets along with daily technical analysis with our Gold Donor program.

Where We Are, Where We're Heading (2013) - The annual 2013 Ticker

Links and Blogroll

Our policy on reciprocal links: Send us an email with your information and why you think your blog or news site would make a good addition - in most cases reciprocal link requests will be granted.

Legal Disclaimer

The content on this site is provided without any warranty, express or implied. All opinions expressed on this site are those of the author and may contain errors or omissions.

NO MATERIAL HERE CONSTITUTES "INVESTMENT ADVICE" NOR IS IT A RECOMMENDATION TO BUY OR SELL ANY FINANCIAL INSTRUMENT, INCLUDING BUT NOT LIMITED TO STOCKS, OPTIONS, BONDS OR FUTURES.

The author may have a position in any company or security mentioned herein. Actions you undertake as a consequence of any analysis, opinion or advertisement on this site are your sole responsibility.

Looking for "The Best of Market Ticker"? Check out Ticker Classics.

Visit the forum to discuss this and other investing-related topics; see the FAQ on the forum for information about Gold Donor status including access to our technical analysis video server.

Market charts, when present, used with permission of TD Ameritrade/ThinkOrSwim Inc. Neither TD Ameritrade or ThinkOrSwim have reviewed, approved or disapproved any content herein.

Market Ticker content may be reproduced or excerpted online provided full attribution is given and the original article source is linked to. Please contact Karl Denninger for reprint permission in other media.

Submissions may be sent "over the transom" to The Editor at any time. To be considered for publication your submission must include full and correct contact information and be related to an economic or political matter of the day. All submissions become the property of The Market Ticker.

Leads on stories of current economic and political interest are always welcome. Our fax tip line is 850-897-9364; please include contact information with your transmission.

User Info	The Scam Of VaR "Risk Models" in forum [Market-Ticker]
Mtdm Posts: 211 Incept: 2009-07-23	It's not a problem with the model itself, it's a misapplication (accidental or deliberate, you take your pick) of it which diverges from its original purpose. It's the adoption of the model into regulations (e.g. governing financial reporting, capital reserves adequacy), which then result in people trying to game the system by structuring crap with all the risk in the tail. And I don't believe that JPM don't have other internal models as well which reach out at least somewhat wider into the tail (four or five nines). No, the bottom line is that it's all about ineffective regulations driving a gaming of the system. One dollar of capital? Yes, please. 2012-05-18 08:49:42 Permalink
Grashopa Posts: 2608 Incept: 2009-02-03	Yes its a problem with the 'model' - it assumes everything is random and there is nothing at all random about it. ---------- Theft is evil 2012-05-18 08:59:28 Permalink
Irishblues Posts: 278 Incept: 2010-12-18 Wisconsin	The vast majority of VaR models rely on market returns being normally distributed or thin-tailed; we've seen repeated evidence that returns in the market are not thin-tailed - and are certainly not normally distributed ... and yet, those assumptions are still used. Most of them also have a lower bound that implies "... but there's no way losses can ever get worse than this" which ignores the possibility - no matter how small it may be - that losses can indeed get worse. Add to that the fact that, as Gen describes above, the vast majority of people don't understand what VaR says, and you have a recipe for disaster. Last modified: 2012-05-18 09:44:12 by irishblues 2012-05-18 09:43:46 Permalink
Rutben Posts: 1414 Incept: 2007-07-27 Phoenix, AZ	I believe the reason Buffet called them "weapons of mass destruction" was because counter parties to the same "transaction" use the same VAR model but have the lattitude to change the assumptions of a variable to suit their own outcome, thereby exacerbating divergence rather than "netting out". After he trapped himself in $35 billion? of these transactions, he no longer talks about this, now preferring to extoll the virtues of bailouts....until he can unwind his positions. 2012-05-18 11:55:41 Permalink
Mannfm11 Posts: 3533 Incept: 2009-02-28 DFW, Tx	Nassim Taleb writes about this in The Black Swan. For one thing, he says the math of these guys is Gaussian, but the reality is something different. 95% in statistics merely means the result will fall within the bell curve presented. I think KD presents this very well. Being they use a day to day risk model, the model has to depend on the capacity to exit a trade within a day, so this needs another model, the risk that the markets will remain the same and liquid. The only thing I can say that almost no one saw coming was the total disappearance of the market for mortgages in 2008. I wrote a post in 2001 that I hope is still out there somewhere. In it, I presented a scenario where FNM, FRE, GE and GM would all blow up when the money markets went awry. It is quite amazing that was the list of companies that blew up, though GE is still in business by virtue of government guarantees. The collateral went to **** and the elephant who walked in the door found it had shrunk to the size of a mouse hole. This is the purpose of the QE's and LTRO's. Between the US and Europe, there is probably $100 trillion in various real debt assets, all of which can suddenly turn to mush in the markets. This is why banks are hoarding cash and why other banks have to have more QE, as they can't manage cash flow pledging the junk collatral they have on their books. In part, this is what the various derivatives in the markets are about. Make believe liquidity. The positions are massive enough that any one flinches and the market disappears or gets so prohibitively expensive that collateral posted sinks even the largest of players. Ask AIG. ---------- The only function of economic forecasting is to make astrology look respectable.---John Kenneth Galbraith 2012-05-18 14:44:28 Permalink
Poid Posts: 610 Incept: 2010-03-08	VaR tells you what the results will be in roughly "normal" times (ie if the near future resembles the near past). It cannot tell you anything about what will happen in the tails. It doesnt matter what distribution you use, what confidence level you assume, the tails in a VaR model are the result of assumptions that simply cannot capture that tail risk. The tails are longer than you have assumed: even if you use a fat tailed distribution you wont be extending that tail far enough to capture genuine tail events. The tail-VaR models were developed to do this...however they are no better in practice (averaging the tail risk instead of taking the 95% or 99% point just gives you a slightly bigger number, but does not model the tails any better and does not extend the length of the tails). The other point about VaR is the better the recent past the worse your VaR model performs as a risk measure, as a benign environment reduces the modeled tail risk. So risk according to your VaR reduces just as real risk is increasing. The model itself pulls your pants down when you least expect it. Lastly if the entire industry is using VaR with similar data as a basis for their activities, it will actually concentrate risk and enlarge the tail risk that players are exposed to, because everybody is using the same parameters to get in or out. Unless a company has strong risk management (across the entire corporation, and that includes having the right culture) a reasonable internal stress testing and scenario analysis methodology, and even better has done stress-to-fail scenarios, they wont understand their own pressure points ahead of time. Without those overlays to VaR you have no idea of the risk you are actually exposed to. Last modified: 2012-05-18 18:36:08 by poid 2012-05-18 18:33:32 Permalink
Schwantz Posts: 5815 Incept: 2007-11-12 Toronto	Quote: But it gets worse: Statistical models are all about random distribution of events. The problem with such models is that markets are, in the "tails", not random at all. The premise that a large enough group of people produce statistically-randomized outcomes is only true until "something happens" and the actions of participants become cohesive (e.g. a bank run) -- then your entire model goes out the window as the very predicate on which it was based is invalid under those conditions Karl I'm glad you're pointing this out. I had this very argument with a friend who is a Berkeley PhD in economics over 15 years ago, and they refused to listen. This grave error is at the core of almost all risk management models, and when combined with the arrogant flawed Keynesianism of Central banking, you get amplified oscillation to the point where you can only control variation by jamming the risk onto a bigger system that is able to absorb it. We've now run out of bigger systems. So yes its a scam perpetuated to preserve existing wealth and power, but it will fail and will take everything down with it, unless people are able to recognize the truth first. Quote: It cannot tell you anything about what will happen in the tails It's all tail. Go translate 'schwantz' from German and you'll get an idea of my dark sense of humor. Lol ---------- When the system is corrupt absolutely you must seek representation by those who are absolutely incorruptible. Last modified: 2012-05-19 10:10:54 by schwantz 2012-05-19 10:05:04 Permalink
Argos Posts: 6305 Incept: 2008-03-23 The Green Mountain State	Lisa Pollack takes a look at JPM's VaR model (and how it was missed) in an entry titled, "Regulator Captured, a Case Study": http://ftalphaville.ft.com/blog/2012/06/.... So it goes when, as Chris Whalen and Josh Rosner put it, the regulators are the lobbyists. 2012-07-01 09:44:13 Permalink
Nanna Posts: 5658 Incept: 2008-01-20 NY State	Not surprising IMO that the big bankers for Enron adopted the off-balance sheet leverage synthetic securities business model. When dealing with large volumes of whatever, leverage can enrich those who are compensated to an extent as a percent of profits. ---------- "There are fluctuations in the market that don't mean anything."Ira Gluskin, February 14, 2012 2012-07-01 12:15:32 Permalink
Argos Posts: 6305 Incept: 2008-03-23 The Green Mountain State	FT and FT Alphaville coverage of VaR magic at Morgan Stanley: http://www.ft.com/cms/s/12b31c9a-1936-11.... http://ftalphaville.ft.com/2012/10/18/12.... (I'm short MS.) 2012-10-19 08:54:31 Permalink

Quote:

Quote: