Is all quantitative financial risk management bunk?

October 28, 2010

The comments to my post here last week on Benoit Mandelbrot were for the most part significantly more sophisticated than the post itself. So, since my days at Chez Felix are numbered, I thought I should avail myself of the brilliance of his commenters while I still can to ask a very basic question: Is the practice of quantitative financial risk management one big con job?

That’s one of the key arguments in Amar Bhidé’s new book A Call for Judgment: Sensible Finance for a Dynamic Economy. He says in the book that the approach to risk management that grew out of Harry Markowitz’s portfolio theory, Bill Sharpe’s Capital Asset Pricing Model (yes, I know it wasn’t just his, but he was the first to publish) and Fischer Black, Myron Scholes, and Robert Merton’s option-pricing model—all of which netted Nobels for their (still-living) creators—is fatally flawed because it depends on predictions about future volatility, and no one knows how to predict future volatility.

The funny thing is that the Markowitz/Sharpe/Black/Scholes/Merton approach arose in part out of the realization that it’s really hard to predict the trajectory of an individual stock—and that even if you did figure it out, others would start to imitate you, eventually affecting the the trajectory of the stock and rendering your predictions invalid. A stock’s future volatility might not be easy to predict, they reasoned, but it was much easier to predict than the stock’s future return. Bhidé turns that argument on its head:

Forming reasonable, subjective estimates of a stock’s return can be a challenging exercise. Predicting whether and by how much IBM’s price will appreciate requires researching and thinking about several factors such as its project plans, relationships with customers, existing and potential competitors, exchange rates, and the strength of the economy. With volatility, because there is no sensible way to think about what it should be, there is almost no choice but to take the “easy way out”: Calculate historical volatility. Shade to taste.

Now the objection one often hears from those in the financial world is that practitioners have moved on from the simple volatility models of the 1960s and 1970s. And that they have. But the models they use still assume that they know how to predict volatility, right? Is there evidence that anybody actually knows how to do that—not just short-term but through an entire market cycle? Or is everybody just shading to taste?

Another issue with quantitative risk models that even many of their creators acknowledge is that when a particular method of modeling risk becomes popular enough, it begins to affect market behavior and creates new risks that the model cannot see. I think this may have been a major contributing factor to every financial panic of the past 25 years, starting with the 1987 crash. (Well, except maybe the dot-com bubble and crash. I can’t really blame that on risk models. Although it is perhaps telling that the dot-com collapse wasn’t really a panic.)

You can’t prove the cause-and-effect, but it is clear that financial risk models have repeatedly broken down after years of seeming success. Which shouldn’t be all that surprising: It is in the nature of financial markets that every good (that is, money-making) idea eventually becomes a bad one. The difference between a momentum-investing formula and a risk-management model is that a risk-management model is supposed to, um, manage risk. And so I guess the question remains: Are all financial risk-management models ultimately a joke?

I really don’t know the answer. That’s why I’m asking you folks.


If I recall correctly, Mandelbrot criticized the risk ‘models’ when they were first introduced, but he was ignored because he couldn’t offer spiffy alternatives that the market players could use, so that they just settled for the bad ones. So basically, everybody (should) know(s) that the models are crap, but they don’t care so long as they can make money (and be bailed out). (And why should they care? The US obviously isn’t a democracy governed by the rule of law any more – it’s just crony capitalism, with a citizenry that is unable to figure out who to replace their political leadership with.)

Posted by Foppe | Report as abusive

I’m trying to understand what exactly your criticism is. The financial crisis wasn’t related to CAPM or option models. CDO models are more to blame. And we had more losses than these models anticipated because they had unrealistic forecasts for home prices and did not adequately account for default risk. People put too much faith in weak models. This doesn’t mean the whole practice is bunk. Just those models and the assumptions of the people using them.

When you’re dealing with portfolio optimization and risk management, the inputs are expected returns, volatilities, and correlations. You can still have an optimal portfolio ex ante if your expected volatility forecasts are wrong ex post. Further, you can determine ex ante how much you expect to lose based on your inputs. However, if your ex ante forecast is wrong, you may lose may ex post. This doesn’t mean its bunk.

Posted by jmh530 | Report as abusive

Justin, the concept of “risk management” is at the heart of my personal investing strategy. But I’m an unsophisticated investor. My concept of “risk” has nothing to do with volatility or short-term price movements. Instead I focus on efficient businesses with a history of strong management, low leverage, and predictable cash flow. It is both complex and subjective (as Bhide points out using IBM as an example), but then I don’t need to be precise. ANY insight that I can add, above and beyond what is reflected in historic volatility, gives me a risk-favorable return. And that is the essential point. Models don’t need to be perfect to be profitable, they merely need to be slightly better than throwing darts.

The fierce competition between quants, all using similar models, leaves them fighting over diminishing scraps of profit and increasingly exposed to flaws in the model. That is good news for anybody who takes a different approach to analysis.

One final point is that the relevant risk factors change over time. Some businesses are sensitive to interest rates, others to foreign exchange, still others to commodity costs. A business that is exposed to a certain risk factor will be more volatile when that risk factor is prominent and less volatile when people are worrying about other things. A big part of risk management is figuring out WHICH risk factors are likely to play a prominent role in the markets.

There was a time when bank stocks were quite staid.

Posted by TFF | Report as abusive

“it’s really hard to predict the trajectory of an individual stock—and that even if you did figure it out, others would start to imitate you, eventually affecting the the trajectory of the stock and rendering your predictions invalid.”

“Another issue with quantitative risk models that even many of their creators acknowledge is that when a particular method of modeling risk becomes popular enough, it begins to affect market behavior and creates new risks that the model cannot see”

These are examples of how a market-driven economy is full of positive feedback loops, and systems with significant positive feedback loops cannot be accurately modeled or controlled. The systems will always oscillate, and that is not the bad thing, as oscillation is unavoidable, and you just want to minimize the frequency and amplitude of the oscillation. The bad thing is the transients that positive feedback loops introduce into the system. They can be called black swans or outliers, but they will happen because positive feedback, when left unchecked, will blow things up. Go put your microphone next to your speaker and see how pleasant it sounds. Or tell your friend (or client) to buy a stock or some other asset because you just made money on it, and it’s going to go up (which will happen when more people buy the stock). It will keep going up until there aren’t enough buyers, and then it will drop even quicker than it rose. Do that kind of trading on the financial system (not just assets like narrow industry stocks), and watch panic ensue.

So if you can’t model a system with positive feedback, it’s impossible to quantify the risk. If your job is to manage other people’s money, maybe it makes sense to turn to the “experts” for their “expert” advice on the risks you are taking with their investments. But don’t count on it being worth much.

Posted by OnTheTimes | Report as abusive

As a practitioner for the last 20 years I’m perhaps, um, biased, but it seems to me that there’s no question that one can build models that do a better-than-random job of forecasting portfolio return volatility. Thanks to fat tails and volatility clustering, naive users of VaR models may be surprised at the frequency of “large” events, but these phenomena are reasonably well understood, if harder to forecast.

It’s clear that no model can predict a true black swan – an asteroid strike, for example, or more locally, a Texas jury verdict bankrupting a AA-rated oil company overnight (Texaco) or asbestos litigation bankrupting a bunch of toffs (Lloyds). But if that’s all the argument amounts to, it’s not very interesting. We’re always trying to manage our lives in the face of uncertainty – the fact that we could be hit by a bus tomorrow doesn’t usually stop us from planning for next week.

Incidentally, the latter two examples are of a type that comes up all the time – there are many investment strategies that amount to collecting premium (bond spread, cds payments, option premium…) a lot of the time and occasionally losing everything. That’s why you diversify. But if you think diversification is useful, you’re already buying a lot of the underpinnings for quantitative risk modeling.

Posted by FosterBoondog | Report as abusive

As an engineer, I have two seperete thoughts on this:

1. As engineered structure get bigger, more complex, with greater consequences of failure, QA/QC, especially during construction, usually increases to reduce the potential for failure. It appears that the financial markets have the opposite culture: as things get bigger and more profitable, the perceived necessity for “deals” increases and the QA/QC of the underwriting process drops. This increases, rather than decreases, the likelihood of failure in the larger, more complex structures.

2. Larger engineered structures with the potential for progressive failure leading to collapse are generally designed to have sufficient redundancy and factors of safety to prevent such a collapse. As a result, major failures like the 9/11 WTC collapse are highly unusual in buildings and that required a much greater event than was designed for to cause the collapse. The financial sector appears to regard increasing complexity as a hallmark of sophistication leading to greater profitability and therefore promotes it while engineering in few real safeguards.

We need to have real building codes for Wall Street firm products and strategies.

Posted by ErnieD | Report as abusive

Banks took a dive. Insurers did not.

Sure, there was AIG, but the review and the administration of that trading group looked an awful lot like the rest of Wall Street. By and large insurers stayed solvent and healthy.

Note that insurers are “old school”–they did not have the sophistication and the deep mathematical training that financial engineers do. In it’s place, they have some 400 years of experience with annuity products and several hundred years of experience with other insurance lines.

The mathematical training is not more than half of the actuarial exams. History and policy and law form the other half. By the time one is done with the exams, the actuary is convinced that he is the only force between an insurer being solvent and a Ponzi scheme.

Why does no one else see that two professions with similar training and a similar focus on risk performed so differently?

Posted by loopguy | Report as abusive

I did some work on volatility models (GARCH-Family) shortly ago, and was surprised how well these models work. As an example i calculated Values at Risk (sure not the ideal risk measure) and Conditional Value at Risk and compared GARCH with other models. It was far more efficient than the usual suspects (historical vola, standard deviation), and one model even bet the ex-post Value at Risk, when measuring the ratio of excess losses and mean VaR.

In my opinion financial risk management is not perfect – what tool is – but even with regard to its failure in the crisis, possibly because expectations where too high – there has been considerable advancement in the field. Risk management is necessarily a business of damned if you do – damned if you dont: If you safeguard against all possibilities, nothing will happen, and your provisions will seem excessive. If you do not safeguard against all possibilities, something will happen eventually, and provisions will be deemed unsufficient.

Posted by owe.jessen | Report as abusive

So much for the “brilliance of his commenters.”

Posted by walt9316 | Report as abusive

Hi Justin,

Here is a question you should think about:

“What is the VaR of a lottery ticket?”

Say I want start a lottery, do the math, determine how many digits I want for a winning ticket, try to determine the size of the lottery, etc.

Initially, I start out small in my neighborhood and require a 4 digit match to win the pot. The first week, nobody wins and I make a boatload of money. The second week, the same. No one wins and I collect dollar bills and begin wondering why everyone else doesn’t start a lottery. The third week, someone wins the lottery and I lose almost all the money I collected the previous two weeks.

That sucked!

So I go back to the drawing board because I don’t like losing money every third week. I decide that now I want to require someone to match 10 digits in order to win the pot. But everyone knows that it is not easy to match 10 digits, so I need to offer a larger pot. Still it is not enough to entice people because dollar bills add up. So now, instead of $1, I charge 1 penny.

Because the pot is so large, I still collect a handsome sum each week from people who do not understand statistics. Things are great. The first week, first month, and even first year go by and nobody wins the pot. In the meantime, I’m carrying bags of pennies to the bank.

Three years go by and, still, no one wins the pot. Life is good!

What is my VaR?

My VaR is a big fat $0. According to the risk models every Wall Street bank uses, the VaR in this situation would show up as exactly $0. This is because the chance of me losing money is less than 1%.

VaR marks the boundary of what a bank would consider a tail event. A 99%, 1-day VaR of $100M means I am 99% sure that I will not lose more than $100M over the next day. Anything bigger than $100M is considered a tail event.

VaR says absolutely nothing about what happens if a tail event actually occurs.

So if you are a bank and you want to keep your VaR low (which they do thanks to Basel), you would be euphoric if you could find a security that, under normal circumstance, had a very miniscule chance of losing any money, but if it did lose money it would lose a lot, i.e. a security with an extremely fat tail. VaR models are completely obvious to such securities. These securities would fall under the radar and allow the bank to load up.

What is an example of such a security? A CDO of course. Bank VaR models are completely obvious to losses that can happen “in the tails” for a CDO. They only capture the normal market fluctuations, which is (as we know) hugely unrepresentative of the true risks inherent in a CDO.

Also, thanks largely to Mandelbrot, we know how to model securities with fat tails, so that really isn’t much of an excuse, although many banks still rely on multivariate normal risk factors *cough*.

So is the practice of risk management a big con job? In many circumstances and at many banks, it certainly is, but it doesn’t have to be. For example, there are coherent risk measures that consider not only the boundary of tail events, but also what happens if a tail event occurs. Using these coherent risk measures instead of VaR would be a step in the right direction for big banks, but then you still have a question of modeling securities. I think those challenges are surmountable though.

So I do not think risk management MUST be a con job, but sadly it often is. I’m not sure how to fix things though.

Posted by EconoDarwinism | Report as abusive

You have asked very fair question, Justin. I was hard on you for you last post; now I have been well-answered.

Unfortunately, it is not possible to answer you completely and coherently in a blog comment. But I should start by saying that I agree with what FosterBoondog has said. The GARCH mentioned by owe.jessen is an example of the sort of volatility modeling I alluded to earlier.

What else should I say? It is a simple fact that volatility estimates are more reliable than drift estimates – if you are using a frequentest approach to estimation. That is not the approach that Bhidé and TFF take; they estimate drift subjectively. There is nothing wrong with this; on the contrary, I heartily approve – for the purpose of making investment decisions. However, not all clients of risk management are willing to accept a subjective basis for risk assessment. The correct conclusion to draw from this is that one should not necessarily use the same measure of probability to make investment decisions that one uses to manage risk. And bear in mind that over long horizions, drift dominates volatility. Long horizons are of material interest in assessing counterparty credit risk.

More generally, this focus on volatility of an individual asset strikes me as wrong-headed. The issue, as FosterBoondog implied, is portfolio management. The important problem here is estimation of the joint distribution of multiple assets; by comparison, we can be quite confident in our estimation of the marginal distributions of individual assets. But given the number of assets in question, there aren’t nearly enough observations available to form a statistically sound estimate of, say, pairwise correlations. And correlations are dismally inadequate characterization of the joint distritubution! So if you are looking for a hobby horse, I would suggest the correlations rather than the volatilities.

My advice is this: read Ricardo Rebonato’s book “Plight of the Fortune Tellers”. I don’t think it has the answers, but it definitely has the questions right. It is brief (about 250 pages) and accessible, having no equations despite being written by a mathemetician. And it contains the practical knowlege of a quantitative trader and risk manager.

And now that I have praised Rebonato to the skies, consider this: Rebonato was head of market risk at RBS when it blew up. Sometimes, not even intelligence, wisdom, experience, and power combined are enough.

Posted by Greycap | Report as abusive

The best example of quantitative risk management being bunk was spelled out by Christopher Whalen:

He talks about credit default swaps being priced off of volatility. It is brain-dead obedience to debunked efficient market theory to price anything off of volatility (usually noise is just noise!).

Its not just Whalen that’s been saying it — some Chinese gentlemen at the Bank of Intl. Settlements came to the same conclusion.

Basically i-banking clowns were writing end-of-world insurance off of data that is almost, but not completely, random.

Douglas Adams could not have made that up. I hope they have begun doing it differently, but who knows?

Posted by DanHess | Report as abusive

DanHess, again I may suffer from sample bias with Mr Whalen but the stuff i do read like this article is nonsense.

CDSes are not “priced off vols”, there is meant to be a mathematical relationship between a risky bond, a risk-free bond and a CDS. Simply risky bond + CDS protection = risk-free bond. Easy right? So why would risk management be an issue:

1) Operational risk is neglected – the prices might be the same but there will be mismatches in collateral payments and haircuts on repo funding. There will also be mismatches in liquidity which will be reflected in the Mark to market prices on the legs of this trade, in particular in a crisis.
2) Basis risk – it is almost certain that there will be mismatches in coupons and maturities of the three legs of the trade. Especially if you factor in settlement differences. Extra especially in a crisis and double triple super-douper especially if people are worried about your solvency.
3) Of course you can do a “perfect hedge” by buying a custom CDS that exactly matches the risk profile from say AIGFP or LEH and then you have counterparty risk.
4) In a crisis, people don’t care about profits they care about losses. So these relationships will not hold. Especially when an asset class gets labeled as “toxic” and you have a Peston or Morgensen screaming about banks with those assets.

Note this has nothing to do with CCPs or lack of transparency or lack of liquidity in a particular product. It has to do with the boring stuff that usually gets ignored – compare compensation for repo and collateral management jobs with OTC derivatives ones.

Posted by Danny_Black | Report as abusive

After arguing with Greycap on the last thread, I’d like to agree with him here. I am no big fan of volatility modeling, but the more relevant problem to the current crisis is correlation modeling. Without versions of the Gaussian copula – which I think is best described as a shortcut around the correlation problem, but too often has been treated as a solution for it – the availability of CDS for junk CDOs would have been much smaller, and when the bubble popped you wouldn’t have had a collapse of AIG. We might have avoided the zero bound entirely.

That being said, I still have two gripes.

First, I’m not as convinced as Greycap that “over the long run, drift overwhelms volatility”, which is a core assumption all volatility models (including GARCH) must make. And I would agree that any fair analysis of the postwar history of financial markets would support that point.

But there are some pretty good indications outside the historical dataset that fundamentals could be very different going forward. Drift dominates volatility…until it doesn’t. And I’m worried that volatility could be king for some time to come – decades even.

Unless we come up with some wondergadget that simultaneously solves the current financial mess, then makes everyone’s debt and trade balance sheets close-to-sustainable across the globe, then lessens the impact of the Boomer retirements, mitigates the inevitability of future commodity shortages, erases the possibility of severe climate change, and prevents the politically destabilizing effects of all of the above, while also keeping international terrorism and Middle East conflagration to a minimum, then the potential for achieving sustainable smooth economic growth in the U.S. will continue to suck for many years and volatility will remain high. And I suspect the good volatility models, used honestly, will basically devalue themselves after adjusting to a few more years of crap historical data. We’re walking a very long tightrope with a broken ankle. A good time to be a pessimist.

Second, I think that the presence of financial modeling does bad things to the attention span and incentives of market actors – especially large firms where the divisions between executives and quants run deep. Behavioralists have covered this one pretty good – moral hazard, collective action problems, etc. – so I’ll leave it at that.

Posted by DaggaRoosta | Report as abusive

In any complex system, predicting the outputs using only the outputs is by definition nearly impossible. Most economic systems that were or are modelled are complex.

the stat tools for risk are mostly focused on price outputs exclusively. The tools lay approximations on top of unknown processes(GARCH variants etc.) This in no way indicates the processes underneath the price.

My guess is that if given random noise from the any natural process, say for example cow farts in a field placed onto a time series, one could find a quant willing to model it, another quant willing rate it and third willing to put it into a portfolio thus making it more efficient.

this excercise provides no insights into cows, digestion, diet or environment, it is just statistically farting about with outputs.

for those interested I run the “black swan” group on linked in. Membership is limited to experts in the field of risk management managing over $100m in physical or fiscal assets and educators.

Posted by Nick_Gogerty | Report as abusive

Reading these comments is interesting.

I think that the elephant in the room in the financial sector is the entire concept that all of these various complex transactions and models can actually be used to mitigate risk at the TBTF level.

In the end, one of the key precipitating elements of the crisis was the insolvency of the parties that were providing the risk mitigation. There simply wasn’t enough behind the “insurance” to cover all of the demand. One of the basic concepts behind insurance is that there is a steady premium stream from insuring unrelated risks that over time provides adequate capital and income to cover a random set of risks.

In the financial crisis, the “insurers” were being whacked by the same “random events” as the claimants, so much of the risk mitigation process became insolvent requiring actions like the AIG bail-out.

As far as I can tell, the financial sector’s models do an abysmal job of being able to understand the overall risk when correlations go to one and asset values plunge across both the insurers and claimants. Until the models can accurately simulate this condition, they are virtually worthless from a societal perspective because they do not protect society from periodic mass bailouts of the ifnancial industry.

Posted by ErnieD | Report as abusive

Risk estimates are only useful if you maintain low confidence in them?

Curious endorsement.

Posted by jpersonna | Report as abusive

ErnieD, there are a number of issues:

1) First off, some 70% of the losses in the credit crisis were due to good, old fashioned bad loans.
2) The real basis of modern financial theory is not really CAPM or gaussian models but rather no-arbitrage theory. Basically it means that people will take profits if they are there and so you can do alot of relative value pricing. The problem is that in a crisis people don’t care about profits, they want liquidity and loss-mitigation, so this model doesn’t work.
3) Banks are pushed by non-commercial considerations to make decisions. The reason most of the SIVs rely on CP funding is that there are cap reg reasons to favour funding of tenor less than 365 days. Imagine if in your example of buildings that construction companies were heavily penalised for using earthquake-proof foundations in building in San Francisco…
4) There are major feedback loops built into the financial system. Mix Mark to Market in with capital requirements and you have a feedback loop. Prices go up, you need less capital and you are making more profit so your equity capital is higher so you can buy more. Prices go down, you need more capital so you sell, prices go down, you make less profits so your equity capital is lower requiring you to sell…
5) There are operational issues. Models don’t always account for margining – again with AIG because when they were written AIGFP didn’t need to post much margin. Also your middle and back office need to be pretty slick, along with the repo and collateral management desk. This are exactly the places “savings” are sought in most banks.
6) There are also data quality issues. You need to know what the traders are doing when they are doing it and you need them to put in accurate and realistic data. For a number of reasons this is not always the case.

All of the above factors come into play before you even discuss the accuracy of the models. Again I think the opposite is true of what is claimed. In most cases quant models are vulnerable to sharp short-term shocks. If you look at the portfolios of most of the famous financial bankruptcies – LEH, Enron Trading, LTCM, AIGFP – they tended to be pretty profitable when the markets snapped back to “normal” and when they could be funded short-term. Even the BSC crap had what would have been manageable losses for BSC in normal times.

Btw, having not dealt much with the Dark Side, I can’t really comment much on the equity models which always struck me more as sentiment driven. I think it was Fischer Black who claimed that equities were stories and fixed income maths.

Posted by Danny_Black | Report as abusive


One of the problems is that the various rules that you mentioned were viewed by the financial sector as obstacles to be overcome in order to have more freedom and profit. It is plainly obvious that in a downturn mark-to-market valuations would force writedowns. However, incorporating a factor-of-safety in your accounting results in lower quarterly profits in the short-term, even though it would significantly reduce the chance of insolvency as a long-term outcome.

The major problem that I have had over the past two years is that we have institutionalized the concept that a fatal error of that magnitude could actually be profitable as Uncles Ben and Timmy will step into the breach and make you whole.

The analogy of financial failures being profitable as long as you can overlook the short-term collapses is a false one. The financial sector didn’t have a problem with using mark-to-market when the markets were rising but immediately cried foul when they were plunging. If they had used conservative accounting when prices were rising, then they wouldn’t have had the mark-to-market blues when prices were dropping.

In building design, buildings will normally function just fine after a windstorm as long as they didn’t fail in the windstorm in the first place. I don’t believe that you would accept your homebuilder telling you that your house would have been just fine except for the fact that there was a windstorm.

The financial sector needs to put in rational accounting that prevents cascading progressive collapse when prices dip 10% or so. Ultimately, banks should be able to survive a stress test of something worse than a sunny day.

In the area where I live, house prices did not go up much during the boom. As a result, it was difficult to get total mortgage (primary + HELOC) values of more than 80% a house’s value over the past decade. As a result, our house prices are almost unchanged with very few foreclosures, no more than normal recessions. There is no general fear that house prices will continue to plunge, so house sales volume is only somewhat depressed from the peak, once again similar to typical recessions.

The financial industry had the option of using that type of model across the board, but elected to use “new math” to justify taking on more risk. It didn’t work.

We are paying the price of a financial system that has been living on the edge of failure on everything from mortgage loan values to CDS financial backing, to pension plan funding. It appears that even simple transactional process items such as assignments of mortgage notes to securities have been blown because the financial sector has been sailing close to the wind in almost every conceivable way, including the historical function of executing basic paperwork.

Posted by ErnieD | Report as abusive

@DannyBlack, the feedback loop you described, like many in the financial system, is a positive feedback loop. Positive feedback is bad, very bad. If prices going up has a second order effect of letting you buy more, which further drives up prices, that is positive feedback. If prices are falling, which means you have to sell more, which drives prices down, that is positive feedback. The terms positive and negative when used to describe the kind of feedback have nothing to do with it being good or bad, it describes the relationship of the feedback to the direction of the output.

There is not a single man-made system that can be controlled with positive feedback, and financial systems are definitely man-made.

Also, you seem to be dismissing much of the crisis to people making bad decisions on loans. It was the flawed models that allowed people to be reckless (I would say negligent) in making those loans. Take away the models that said mixing up bad loans with good loans like vegetables on a child’s dinner plate will make safe bonds is what caused the problem.

Posted by OnTheTimes | Report as abusive

ErnieD, 10% margin of error would not have been enough.

Here is the ABX in the later half of 2008. It has halved in a couple of months: s/2009/02/abx.jpg

Thats on top of it losing nearly 50% running up to the final crunch. Are you suggesting a bank should run its business based on the assumption that its assets could shed 50% of their value in any one quarter? Or that liquidity would completely dry up? When you are building near Yellowstone do you build with the assumption it nears to survive Yellowstone blowing? How many buildings coped well with Katrina in New Orleans? The majority of investment banks DID survive major crises – such as 87 crash, early 90s recession, 94 peso run, 97 asian meltdown, 98 LTCM meltdown, dot com crash and 2002 accounting scandals. They also looked like they would have survived 2007 subprime until the money markets froze post LEH bankruptcy, which was the financial equivalent of Yellowstone erupting.

I don’t have an issue with mark to market. I believe it is the best measure of what an asset is worth at that particular time. Doesn’t matter that if you hold it for two years you’ll make a big return if you are in a situation where you might need to sell today. The positive feedback loop comes from the interaction with credit ratings, the repo and CP market and capital regs. If you want a more stable banking system more thought needs to be given to how you break those loops – currently none has. Furthermore, people have pointed at things that most certainly did not cause problems – such as defaults on the CDO tranches insured by AIGFP.

I think you misunderstood what i was saying about mid-term vs short-term. I was simply stating that most of the trades based off arb pricing and hedging eventually go back to being profitable because there are relationships between the values. The price differentials may blow out short term but sooner or later – and later maybe TOO late – price has to come into line with value.

As for sailing close to the wind, I think I made my view clear on the foreclosure “crisis”.

Posted by Danny_Black | Report as abusive

OnTheTimes, no argument with me about those feedback loops. I just wish people had spent longer focusing on this and trying to solve these problems than screaming “fraud” or blaming investment banking for the crisis – despite it overwhelmingly being a pure real estate loan issue.

I doubt the originators went into much analysis about loans, let alone used “sophisticated models”. I would point out that contrary to what seems to be conventional wisdom it was not a case of originators having a large supply and sneaky banks selling it to unsuspecting investors but more that there was a huge demand for these loans as raw material for investments. So much demand that the originators simply couldn’t make enough of them.

By the way you CAN mix up good loans with bad ones to make safe bonds. I buy 99million of goverment bonds and 1 million “bad loans” at 50cents on the dollar and you buy the cashflows for 100million. Thats a pretty safe bond. I am using an extreme example because I want to make the point that you can expect a loan to default and you can still have a safe investment on it.

Posted by Danny_Black | Report as abusive

At least part of the feedback cycle was the widespread use of leveraged investment in various guises. When you are leveraging by 30x, it doesn’t take much to wipe you out.

Liquidity problems are comparatively easily solved (government intervention) if that is the only concern. And liquidity dries up in ANY crisis, so that kind of intervention may always be required?

Posted by TFF | Report as abusive


It halved because it was too high to begin with but nobody was accounting for that. There was no margin for even a 10% drop, so once it started to go, it collapsed due to positive feedbacks.

Personally, I sit down periodically and review what the valuations of my portfolio would be at typical bond interest rates and equity PEs (Shiller’s 10 yr real PE). THAT is what I base my long-term projections of 2% real portfolio returns (@ 6%-8% nominal).

I don’t believe that my current portfolio will return a nominal 8% over the next 20 years based on the current valuations. However, pension funds are currently using long-term return values of 8% in their calculations which is why we are probably going to see those implode as well in the years to come.

Regarding your New Orleans example, the federal and local governments made the same error that they made with the financial markets. They took their eye off of the purpose of the levees and how important they were. As a result, a combination of wishful thinking, politics, incompetence, graft etc. resulted in those failures. There was nothing about the levee failures that regular engineering couldn’t have predicted or good construction and maintenance practices prevented.

On the Yellowstone example, that caldera has a failure rate of substantially less than 1 in a hundred years. However, Wall Street has major financial failures about once every 20 years or so where at least a couple of asset classes get hammered. We have had two stock market collapses in 10 years with one of them coupled with a credit crisis. that follows the 1980s S&L crisis, the bear market that culminated in 1982 with double-digit interest rates, the collapse of Europe and Japan in the early ’40s, the Great Depression and on and on…

If financial firms are not structuring their businesses to survive during one in 10-yr to one in 20-yr events, then they should go out of business when that event hits. The mere action of structuring their businesses around surviving that type of event would likely prevent the event from occurring in the first place. We saw this in the complacency that allowed for the Glass-Steagal Act to be repealed in the 90s.

Posted by ErnieD | Report as abusive

TFF, liquidity provision in a crisis is what the Fed system was created to do after 1907. As for liquidity, one can see the snap back relations between different products post the intervention in October 2008.

ErnieD, the ABX went down by nearly 50% prior to the major meltdown in the second half of 2008. Yes you can sit the bubble out – like JP did before getting bought for a relative song or like Phil Purcell did at Morgan Stanley before getting sacked.

I know exactly zero about construction. I am maybe assuming that even the best house in the world is going to be in pretty bad shape after being under water for a while. The point I was trying to make is that no matter how well you engineer there is going to be something that can wipe you out.

As for failures, lots of industries have high failure rates. Look at airlines, or engineering companies, look at US auto companies that need a bailout on a regular basis. The 2008 crisis was not like the other crises of the 20th century. I would say there were two instances in the last 100 years where the financial system in toto came close to collapse. Most of the major banks in 2008 were major banks in 1998 and 1988 and 1978, some got wiped out most didn’t and the failure rate is far lower than say airlines or hi-tech firms. The banks had a pretty high survival rate.

I have to say I don’t understand this fetish about Glass-Steagal. Again most of the losses were caused and borne by boring old lending and commercial banks. Did Glass-Steagal stop Citibank nearly going bankrupt in the early 80s over third world lending and again in the early 90s over credit to poor credit risk individuals?

Posted by Danny_Black | Report as abusive

Danny, I agree that the originators didn’t think much about models or anything else, but they would not have been able to write so many stupid loans if there was no market for them to re-sell. And the market existed only because of a model that said they could be protected by blending loans of various risks. That might have worked if they were able to accurately assign risk, but that wasn’t the case.

Your example of 99% govt bonds and 1% garbage is not fair, as that is not what was happening. If you are correct that 70% of the loans were bad and should not have been made, then there would be no amount of “safe” loans to blend with them, since they would outnumber the safe ones by more than 2-1.

Posted by OnTheTimes | Report as abusive

OnTheTimes, actually that was the point of my example. It seems to be an assumption that RMBSes and CDOs relied solely on diversification to avoid losses which is untrue. Let me give you another example – again admittedly extreme. Lets assume 50% recovery ie that after all costs you get 50% of face value of the loans. Let have have a 100 million of bonds which payout the first 20% of any cash flows. Thats also a pretty safe bond – lets assume for a moment it doesn’t take 16 months to foreclose…

One other thing about quant models and diversification. One of the reasons that correlations are higher than expected is BECAUSE of diversification. If you have enough people buying asset A and asset B together because historically they were not correlated or were negatively correlated then by that very fact they become positively correlated. I can’t remember who or where i heard this story about about LTCM but apparently one of the senior guys was complaining about how LTCM was diversified and how all their assets should not have been correlated. The person who was listening lent forward and said “You don’t understand… **YOU** were the correlation”.

Posted by Danny_Black | Report as abusive

loopguy, just ask Equitable Life and names of Lloyds of London from the 80s how they are doing.

Posted by Danny_Black | Report as abusive


As you point out yourself: “**YOU were the correlation” is the key.

The statistical distributions and models for new investment concepts are based on low inter-connectivity and correlations. However, Wall Street operates as herds. Somebody comes up with a good idea that works when nobody else is doing it. However, they are slow to understand that it will behave very differently when everybody is doing it.

As a result, Wall Street piled up the kindling, started up the wind machine, and threw the match in and then demanded to be saved by the firemen when the fire raged out of control. Nobody made them create complex securities. Nobody made them do poor due diligence on the securities. There were essentially no external forces. On top of that, they lobbied very hard to rid themselves of rules that would put some artificial barriers in place to keep companies smaller, make them less interconnected, and reduce risky activities.

Wall Street deliberately chose to build a large, complex, inter-connected machine that was increasingly dominated by just a few key players where the instability of any one player could take everyone down. Citigroup could virtually go bankrupt in the 1980s without threatening to take down the world’s financial system. The same is not true today.

We are doing far too many bailouts across multiple industries because of perceptions of importance of individual companies and for political reasons. We need to have a financial and economic system where companies can be permitted to fail if they want to. They should certainly not be rewarded for failure which is what happened over the past couple of years.

If we are going to be doing bailouts, then there need to be management change-outs and serious haircuts for counter-parties. For example, the AIG CDS payouts should have been far lower than 100 cents on the dollar. If that caused grief in the counter-parties, then they should learn to not put their eggs in just one or two baskets.

On the civil engineering side, we are seeing classic examples of this in some of the major floods over the past couple of decades. Many small communities are building levees and building floodplains without anybody looking at the system-wide impacts. Meanwhile, they expand impervious surfaces so that more runoff occurs. This gets done along the full length of a river system, In the end, you get more runoff with fewer places for the water to go with the end result of more flooding. In some of the recent Mid-West floods, USACE didn’t even have many of the levees mapped, never mind hydraulically modeled. The new levees were built suing old, inadequate data sets and models, just like the financial system.

The United States society-wide seems hell-bent on creating large, complex, fragile systems with the expectation that bad things won’t happen because they are thinking happy thoughts. Much better understanding of risk and risk management is required across the board. there is less glory in building smaller, less connected stable systems, but the greatly reduced chance of massive collapse is a pretty good trade-off in the long-run.

Posted by ErnieD | Report as abusive

ErnieD, again I have to disagree.

1) Most of the quant driven trading models are based on correlations and relationships between products. Some of it is recognised as being driven by market microstructure.

2) I seriously doubt Citibank – one of the largest banks by assets at the time of crisis in the 1980s – could have gone down without major consequences. Remember that there have been plenty of panics kicked off by the failure of one bank prior to this. Off the top of my head there was the panic of 1907, kicked off by the failure of the Knickerboker Trust. The great depression wave of bank failures kicked off by the failure of Credit Ansalt( possibly spelt wrong ). Interconnectivity and feedback loops are a feature of the banking system, who by definition have illiquid assets and liquid liabilities and so are reliant on the ability to borrow from each other. When that liquidity dries up there is relatively little a bank can do.

3) Which leads to all the fuss about the Fed. The reason the Fed was created was specifically to be lender of last resort in the aftermath of 1907. That was actually its job before it got saddled with somehow targeting macroeconomic stability.

One needs to remember what happened in October of 2008. The CP market froze. Money market funds broke the buck. Inter-bank lending froze. T-bills had a negative interest rate. Banks cannot survive long without liquidity, no matter what their leverage, what their investments, what their connectivity. A risk model of any bank in those circumstances says the only thing to do is file like LEH. This is an attribute of what banks are and do.

To answer the original question posed. In the midst of the crisis, the trading losses were such that I would say most of the quant models held up pretty well but as you can probably guess i am biased.

Posted by Danny_Black | Report as abusive

ErnieD, it is a bit off-topic but since you brought it up…

“I don’t believe that my current portfolio will return a nominal 8% over the next 20 years based on the current valuations. However, pension funds are currently using long-term return values of 8% in their calculations which is why we are probably going to see those implode as well in the years to come.”

I agree, to a point. If you figure inflation at 2%, then nominal 8% returns on those pension funds are likely impossible to achieve. I figure 4% real returns on my own portfolio, and that is invested somewhat more aggressively than your typical pension fund. Moreover, I’m probably wildly optimistic in that half-baked guess.

But how does the picture change if we average 7% monetary inflation over the next 20 years while public-sector wages (comparatively) stagnate? Might that be sufficient to meet the 8% target?

Inflation won’t solve all of our problems (decades of high inflation create an environment that is broadly hostile to borrowing), but I don’t see how we are going to fund the public pensions if it doesn’t materialize.

Posted by TFF | Report as abusive

TFF, you are not. Pension plans are bankrupt and you are seeing a slow default over time, mainly by upping the retirement age.

Posted by Danny_Black | Report as abusive

I’ve written about this before, on this site, and what I had to say then is germane to this discussion.

This has never been about the statistics of risk. It’s about non-linear dynamical systems, a topic the “quants” seemed to have slept through – if they ever familiarized themselves with it at all. Or perhaps they chose to ignore those nagging doubts in the back of their minds, or they just don’t care.

Because that’s what the marketplace for all these exotic instruments is, an iterated discrete non-linear dynamical system, one with lots and lots of feedback. Such systems are capable of transitioning from one state to another in a flash – as in flash crash. Looks just like a statistical blip – the so-called black swan, and the time series look stochastic.

Instead, what you have is a dynamical system that orbits around on one portion of the state space, but it doesn’t have to stay there. The system can transition to other portion of that space where it will do its dance for a while before slipping back onto another wing of the so-called attractor.

My concern is that with the introduction of networked computing power, we’ve simply speeded up the whole process, to the point where collapses such as the one we’ve experienced, will happen more often.

My challenge to Mr. Fox, is to walk over to the Harvard physics department, and to find himself a dynamicist and discuss this with her.

If this is what we’re dealing with, statistics are of little use. I won’t go into why except to say it’s the so-called butterfly effect. Using statistical models to bound the variation in a system that has any endless number of state-transitions embedded almost everywhere is bound to fail.

Posted by NCimon | Report as abusive

Also here is an article from the author of the second best finmath book: elRisk.pdf

Deals briefly with the real issue with the gaussian copula, also there is a pretty good book: isis-Journey-Correlations/dp/0470665661/ ref=sr_1_1?ie=UTF8&qid=1288711933&sr=8-1

Overpriced but short and sweet.

Posted by Danny_Black | Report as abusive

Post Your Comment

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see