Paradise Lost and Regained - an Interlude into Risk Management

Paradise Lost, and Risk Management Regained

mjdudziak

Paradise Lost, and Risk Management Regained Martin Dudziak, PhD Patterns and Predictions Boston, MA The furor over what and why for the 2008 crash of several major investment banks, brokerages, and insurers is in full swing and the punches are being thrown in many new directions and somewhat all at once. Executives and seasoned brokers are being blamed for not seeing what many feel was an obvious, common-sense, inevitability. Greed and lack of foresight within the fuzzy middle layers of brokering and sales, particularly in the domain of mortgages and mortgage-backed securities, is held to the light as the main causus terrible by some. What is probably a “first” phenomenon in the popular press, and among some pundits, is the blame and invective being hurled at the “quants” and in particular the mathematicians, the physicists-turned-analysts, the geeks who came up with overly complicated, obtuse, unintelligible, and yada-yada otherwise “bad” quantitative forecasting and risk management systems. Value at Risk is held by some to be something not far removed from advocacy of drug use or premarital sex. It’s natural for the finger-pointing, and even the axe-throwing. This comes with the turf, it is simple human nature. But eventually, there is a need to look for systematic and rational understanding that can lead to both a way out of the mud pool in which we find ourselves and a method for avoiding the same slip-ups and trip-ups in the future. Making an effort in this direction is the point of this article. I want to see if some things can be clarified, I want to take issue with some stances and advocates, and I want to offer a practical approach that (be forewarned in advance!) involves mathematics, algorithms, and software. However, right from the start let me point out clearly that the proposed solutions and “tools that really work” are not based upon a rigorously “quant” approach to the numbers that go along with the securities (stocks, futures, options, swaps, bonds, whatever) but make use extensively, fundamentally, with the sentiments, moods, trends, attitudes, and expressed intentions of very large numbers of people who don’t talk only in numbers but in very qualitative expressions through text, through words, and nowadays through words that are more often than not close to real-time and usually online. Having given some sketchy advance warning that this is not yet-another-VaR in the making, I will proceed, by making a running commentary on an intelligent and provocative article that appeared recently in the New York Times, “Risk Mismanagement,” by Joseph Nocera (January 2, 2009). 1 My thesis is simple. There has been tremendous risk mismanagement. It is not as simple as a decade-plus long adoration and infatuation with Value at Risk (VaR) methods, it is not as simple as saying that people were blind to the 1% outlier, the “Black Swan” (although there is much that has to do with not wanting to see or even consider such anomalies and low probabilities, and the herd instinct and meme behavior that keeps people mesmerized, sedated, and generally programmed to avoid such things), and it is quite scarier, in fact, because of the multiplicity of nonlinearities and black swans, that we are facing today, particularly, in 2008-2009-plus. The good news is that there are tools, not magic wands, but very sensible, tested, validated tools, that can help us and not only in the securities markets but in 1

“Risk Mismanagement, by J. Nocera, New York Times, 1/2/2009. The article is available online at http://www.nytimes.com/2009/01/04/magazine/04risk-t.html?_r=1 (this article is denoted as Nocera[1] in future references)

1

Paradise Lost, and Risk Management Regained

mjdudziak

all walks of economic and social life. The big question is whether or not people and institutions will respond to what is being offered straight-up as a good tool to use in this kind of inclement weather. Few will dispute that what happened was an unraveling of global financial markets that was linked with a number of unanticipated “1%” events coming into being, all around the same time, and many linked with a class of securities that were very vulnerable to the unexpected and nonlinear happening. Value at Risk (VaR) systems told traders and managers what was likely and not likely, based upon the constraints of the data going into the analytical systems. This did not go back very far in time, nor would it matter to try and go back to the 1970’s or earlier. The contexts of all markets and trading, and home ownership as well, to not forget the mortgage market, were all qualitatively changed as we moved into a fully global marketplace, with a population of 6.5 billion, and an internet in the majority of homes, offices, and pockets that was simply not there, hardly anywhere, even ten years ago. VaR did not predict what, how, when, or if the “1%” risks would occur, or what could be the implications, the connections and consequences to other events. There was very little modeling, much less thinking, about these interconnections, dependencies, and cause-effect relationships. what happened was that a cascade of dominoes began to fall, and those falling lines led into other branches and unexpected directions, but ahead of all this, for years preceding 2008 and the beginning of different domino cascades, there was very little going on within the “quant” departments, or in the board rooms, that showed how tightly coupled everything had become, and how the dominoes were lined up in ways such that a “1%” could do what in fact has become fact. Nassim Nicholas Taleb, author of “Black Swan,” is quoted as saying in a lecture at Columbia earlier last year (2008), “Why do people measure risks against events that took place in 1987?” he asked, referring to Black Monday, the October day when the U.S. market lost more than 20 percent of its value and has been used ever since as the worst-case scenario in many risk models. “Why is that a benchmark? I call it future-blindness. If you have a pilot flying a plane who doesn’t understand there can be storms, what is going to happen?” he asked. “He is not going to have a magnificent flight. Any small error is going to crash a plane. This is why the crisis that happened was predictable.” I will get back to this point, but it is critical to also understand that, in keeping with Taleb’s analogy, the weather system is also changed. Storms and ordinary weather systems are not what they have been in the past. You could say, we have had “global climate change” in the markets, presaging that in the physical world. A major problem with how economic performance is measured, across the board, is that references are made to volumes of trade, volumes of business activity in general (productivity, GDP, unemployment claims, etc.) in comparison, today and x years back, but x years ago, there were significantly fewer people (population), workers, students, consumers, etc. National budgets were lower, all sorts of things were markedly different, including popular expectations, impressions, knowledge of the markets, and abilities to react and change things on virtually a moment’s notice. All this does not get accounted for even if one is adjusting for inflation or population numbers over some years or decades. Thus, the resulting comparative numbers and percentages are doomed to be misleading, and we do not need much misdirection when it comes to lines upon lines of dominoes in a crowded space to generate consequences that were not only unexpected by the number, by VaR or any other conventional statistical approach, but inconceivable, as just not something that could happen. The analogy with flying and storms is very good. But think of this, too. Much of how we learn, when it comes to expertise, to fine-tuning, to conducting some process that is physical or mental, in sports, in driving, in sailing, in flying a plane, comes from experimenting with boundary conditions, with limits. Just watch the behavior of babies, and then children as they begin to walk, to do things with rules and behavior at home, with riding a bicycle. The ace test pilot pushes himself just to the very

2

Paradise Lost, and Risk Management Regained

mjdudziak

limits, and early on, with a new plane. If he is really an ace, he knows where to stop, where to hold back before things no irrecoverably beyond limits. Consider the storms in the sea or the sky, for sailors and pilots. If everything is done by the book, by the quants, by the experience of sailing/flying and not having anything go intensely non-linear, then a person can be very skillful, smooth, successful. They can sail nicely, even win a race. They can fly around and maybe even do acrobatics, or fly smoothly and the passengers are all happy for the quiet ride. But if they do not know how updrafts and downdrafts can rip a plane apart or send it into a tailspin, if they don’t know how to sail into a wave and avoid a 12’ broadside that will capsize the boat, then they are pretty close to doomed from the get-go. This is precisely what Taleb and others point out with respect to the established market players. “People tend not to be able to anticipate a future they have never personally experienced.” 2 This is true, but it can also be said that people tend to block out the anomalies and indicators, even when they are “loud and clear,” which give warnings and presage the coming of something that will either be very rare, very much distant, from long ago, or very new, but in any case still predictable that “something weird is coming down the pike.” This is a matter of psychology, both individual and collective. Now think again of local vs. global minima, and the varieties of local minima you can find in a landscape. Also about catastrophe functions (Thom variety).3 And shift to “regular ordinary life in the home or backyard.” There are plenty of situations where one has never experienced X but something is going on, and there are indicators, predictors, that cannot tell exactly what is X and how it will unfold (like a topological transform, a surface change), but we certainly know that something drastic is ahead or around the corner. We can call it intuition or second-sight sometimes. It is not magic but more like common sense. See the figures on the next page. They are unordered, just here as examples. But bear in mind that if you are on a roller coaster at the amusement park, and you keep going up and down, up 30’, down, and up 40’ and down, and up 30’ and down, and up 50’ and down, etc., and you have even an iota of a brain, then you certain can tell, and your body anticipates it definitely, that when now you are climbing, climbing, climbing up higher than before, 60’, now 70’, now 80’, and still crawling upwards, that you are going to soon, sometime, but definitely Eventually, be In for A Really Big One. This simplistic intelligence which surely even a cat or dog will have (of course it would have jumped off the coaster a long time ago), is what all the quants and traders on Wall St. seem to have been lacking. But wait a minute! They are intelligent, and have all sorts of tools, right?

2

Nocera [1], p. 4 The word "catastrophe" has the connotation of an undesirable and disastrous outcome but this is not always the case with such functions. Consider an epidemic disease like the flu or smallpox. While the explosive onset of many epidemics is certainly a negative event, the same mathematical process operates but in reverse when an epidemic disease virtually disappears suddenly from an previously infected population. How does this occur? The balance of susceptible and immune individuals shifts from the proportion minimally required to sustain an epidemic to a marginally smaller proportion where the probability of transmission of an infectious agent from an infected host to a susceptible host falls below the critical level required to sustain the epidemic. 3

3

Paradise Lost, and Risk Management Regained

mjdudziak

Now here we come to something interesting, not well publicized. Something not that uncommon in other fields of anomaly/pattern detection pattern recognition and classification, in the biological and physical sciences, in signal processing, but it seems very uncommon in the conventional and established financial markets. I think a big reason for this discrepancy is due to simple aspects of human psychology, patterns of behavior that most people do not want to admit exist in their circles or within themselves. Emotions of greed, fear, and the herd instinct. Unconscious but powerful urges of not wanting to hear bad or disturbing news that makes one have to think, reassess, and even walk away from what seems to be a hot, going thing. Taleb seems to miss this dimension, and perhaps it is from also being too much immersed in the world of the established numbers, the charts, the stats that have been accepted as being the primary thermometer of how well we are doing as an economy, and more broadly, as a society. “VaR can be useful,” said one of the risk managers. “It depends on how you use it. It can be useful in identifying trends.” “I think VaR is great,” said another risk manager. “I think it is a fantastic tool. It’s like an altimeter in aircraft. It has some margin for error, but if you’re a pilot, you know how to deal with it. But very few pilots give up using it.” 4 Taleb appears, from his writings and speech, to be hell-bent on throwing it all away as useless, or so it seems from my vantage point. Putting all or simply too much of the blame upon VaR and those who used it as a basis for many years of comparatively successful business management and growth may be leading us today to look upon or sector or group as the cause, giving us impetus to jump to a new set of conclusions, a see-saw reaction, a ying-yang effect, and meanwhile, we may be missing something that has been missing from most of the quantitative analytics, pattern recognition, and forecasting systems for years, namely the important and defining role of people and their sentiments, their opinions, their attitudes, and their expressions of confidence or lack of confidence in what is around them in the markets. Part of my thesis here is closely coupled with this matter of noticing when and of what type are the shifts, coming from popular sentiments expressed both inside and outside the conventional trading arenas, that can provide a “heads up” that the everyday models, the tried and trues, like VaR, need to be supplemented because there is something, however, fuzzy, that is “coming down the pike and around the bend” that has not been in those models. So I am working toward the view that we do have indicators, like a change in the weather that is not an immediately measurable increase in wind speed, or cloud color or density, or even today’s temperature, that can warn us to change. Sailors are good at this sort of thing. At least good sailors who come back. I am not merely working toward a conclusion in the abstract, nor a pointing finger without something concrete that can be used to improve the situation. That, a set of tested and proven tools, will be described later, and it has very much to do with massive data, needles and also horseshoes in the haystack, and human sentiments that can be dis-covered. Certainly VaR and other techniques can be useful, as indicators of local trends and local (short-term) dispositions. VaR is demonstrated as a way to identify indicators of what can be termed mainstream, 90-99% trends that are also what one can expect most people, most of the time, will accept and use to operate. But the first step to both success and failure is one of recognizing the boundaries, the limits of the model. It’s like conduct in a military tactical operation. If you know that the enemy army of 4

Nocera [1], p. 4

4

Paradise Lost, and Risk Management Regained

mjdudziak

10,000 has to be walking through a certain valley, you can make some pretty sure-fire accurate predictions about where they will walk and not walk, where they’ll likely cross streams, etc. But you cannot use that to base your entire plan 100% and not consider that a very smart and “thinking outside the box” opponent won’t have something up his or her sleeve, like sending part of his force over a very difficult piece of terrain in order to get behind your lines. (This was done repeatedly by the brilliant Stonewall Jackson and also Nathan Bedford Forrest during the Civil War; George Meade was one of the know-it-alls from the Union side who repeatedly fell into traps; maybe he was reincarnated as a manager in a certain Wall St. firm, who knows?) So what can we do that will be astonishing and striking like a blow by Forrest’s mountain cavalry? What kind of guerrilla combat might we wage with mathematics, software, and intuitive, commonsense reasoning that can tell us when it is time to stop business (forecasting) as usual and that we are not in a steady-state world, nor in a linearly expanding one either? We can focus upon three categories of information that are widely available and accessible, two of which are easier and more straightforward as far as source availability and the detection process. First there is that which is not an anomaly but clearly evident. Assimilation of explicit dialog and expressions of opinion – Sentiment-Tissue – from the organism of global mass society. What people are saying, thinking, feeling, unhidden, unsecretive, out in the open, and mainly through the web communities, the social networking. This is what is explicit. Second there is the unhidden but implicit data, the actual consumer behaviors, but not necessarily those that any of the conventional economic models are looking at, or rather, in addition to those. Besides housing sales and prices and new home construction starts, there is something else of great importance, and it seems to have been constantly overlooked, but the data is there! It is quite public and not so hard to get (although it can require purchasing). This includes (sticking to the housing example for simplicity): • • • • • • • •

amount of time for existing homes put on the market to sell numbers of offers and periods of negotiation on a contract success/failure rates for mortgage applications relative to particular properties and neighborhoods ratios between asking prices and selling prices changes in average features (like room numbers, sizes, etc.) in new homes built or ordered changes in types of homes being sold (e.g., townhomes, condos, single-family, duplex, etc.) changes in real estate agent economic indicators (harder to get at this but maybe not if numbers removed from identifying names) and there are other indicators that can be found

Third, there is the anomalous, outlier, asymmetric data. This is what is so ignored. It is harder to find, it is rarer, and it may be hidden both intentionally as well as unintentionally. We are looking for things that “stand out” as being NOT what one would expect if everything is going to stay long-term, or long-enough, in the same basic “surface” (topology, again). We are looking for the equivalent of a phase-shift in matter, or you can think of it as a disruptive surface change. So for example, picture that you are on a curved surface. You can see and measure curvature, positive and negative. You can see a dip and also see the rise beyond it. You can operate as one will with a good instrument set onboard a plane. But then you look in one direction, and off in the distance, there is nothing. There is only a curvature downward, but you cannot see the return-rise ahead of it.

5

Paradise Lost, and Risk Management Regained

mjdudziak

This should tell you something. If you have ever gone hiking in the mountains, especially in the snow, you know where there might be an overhang. You “sense” that if you step There, you may step thru the snow into – thin air. People who cannot perceive and think this way, if there are in the mountains in the snow, they are likely to get killed. On Wall St., they got bonuses and reinforcement to walk even closer to the edge of the snow cornices. There is a big part of your problem. As prelude to describing the better approach, one that could have helped some firms and large numbers of investors to avoid the Massacre of 2008, I want to show a few illustrations about functions and the surfaces they represent. Below are four simple pictures that illustrate a poorly-understood and often ignored branch of nonlinear mathematics known as catastrophe theory. The French mathematician René Thom 5 is generally credited as being one of the pioneers in this field. In terms of what I am trying to convey to the reader, a picture really is worth at least a thousand words, because the nature of what can happen is evident from the first glance. Speaking of strange attractors and the difference between a chaotic vs. a non-deterministic complex system, such talk will not resonate with people who have experienced losses in the billions almost overnight, or devaluation of a home property that is one’s life savings and financial security to being less than the payout on the mortgage loan after fifteen years of steady monthly payments. But a picture may have better outreach effect. 6

5 6

ref to R. Thom’s work on catastrophe theory for image sources, refer to endnotes

6

Paradise Lost, and Risk Management Regained

mjdudziak

There are two very important points here about these four illustrations. First, they represent functions that map to many different situations in the physical, biological, and social worlds. They are not uncommon. Simply put, there is a lot in life that behaves this way, including processes that affect financial markets because they affect the ways a lot of market-influencers behave. Now it cannot be proven that the financial markets of 2008 or 2009 are at a particular coordinate on such a surface. The argument here is definitely not for making a predictor based upon a new and improved VaR that assumes any specific behavior. Doing so will lead to the same problem – over-reliance upon a set of quantitative forecast numbers that will ultimately lead to decisions that do not take into account the unexpected and unpredictable in terms of events, and then we will be back in as vulnerable a place as the hedge funds and investment banks have put themselves over the last few years. Secondly, when there is a change, it can be dramatic. One minute you are in a positive curvature, then suddenly it is negative. All along, for a long time, some values are increasing in a manner that seems to be the way things are meant to be, ought to be, and will continue to be with no evidence of going abruptly in a new direction, and then suddenly everything is “upside down.” For an ant crawling around on a catastrophe function surface, this is not much of an issue. Have sticky feet, have no problem. For people and their investments, it can be earth-shaking-shattering, like – falling off a cliff. I want to return to Nocera and his article, for a moment, because he makes some particular good points, not in defense of VaR but in some way balancing things out with the strong invective from Nassim Taleb. I believe that this sort of balance is not merely good style or practice in itself, but it paves the way for discussing how a set of techniques, embodied in a well-known but underutilized set of algorithms, can be an aid to those of us who need or want to walk perilously close to the edges of cliffs and still not go off into oblivion like most of Wall St. did last year. “But a computer does not do risk modeling. People do it. And people got overzealous and they stopped being careful. They took on too much leverage. And whether they had models that missed that, or they weren’t paying enough attention, I don’t know. But I do think that this was much more a failure of management than of risk management. I think blaming models for this would be very unfortunate because you are placing blame on a mathematical equation. You can’t blame math,” he added with some exasperation. 7 Good points, balancing out Taleb but also those who simply did not Think. Now what can we do that does not lead to a situation of dependence upon a quantitative model, no matter how sophisticated, even one designed to look for “black swans,” that is going to result in a high probability of missing the new anomaly that has a different set of numerical characteristics, one that slips by the filters and detectors? Remember – since we do not know in advance what are the parameters that may go “off the deep end” and pull down the “house”, we will always have a challenge in trying to set up a closed system (model) that relies upon a finite class of inputs (e.g., market values). We don’t know, and cannot know, all of the forces that can drive a topological change like we see in the general catastrophe function type. The best solution lies in expanding the data set of observations, the field of vision. It’s just like within conventional optics, or sonar, or radio telescopy. The larger the array, the more pixels per image, the more detectors per square mile, the better resolution can be obtained. What this translates to in terms of identifying high-momentum market-moving anomalies and trends is not something about images or sonar but about different data sources for gaining heads-up information about behaviors in buying and selling of large numbers of goods. This is why we need to be looking closely, more than ever, at the 7

Nocera [1], p. 6

7

Paradise Lost, and Risk Management Regained

mjdudziak

information governing that buying and selling, not only in the actual transaction-space of goods, real estate, and securities being traded, but in the potential transaction-space where people of all sorts are expressing their intentions and especially their dispositions and inclinations. This is where the use of techniques to identify mass sentiments, the moods and predispositions of large number and variegated types of consumers, comes onto center stage. Borrowing from a comment made over a decade ago in the context of the ill-formed Project Genoa, a precursor of the DARPA/ARDA/DTO project TIA (Total Information Awareness), “Instead of running off with huge magnifying glasses and binoculars ‘looking for needles in haystacks’ (John Poindexter’s favorite quote) we need to be identifying who’s talking about trips to the barn and who’s seen walking around with pitchforks or straw on their shirts and boots.” In other words, there is a huge amount of information that can be right on the surface, easy to obtain, “in our faces,” and it all has to do with behaviors, comments, and sentiments. When we are able to learn these in the context of what large numbers of people think and feel about the economy, we have something to work with for predicting how a lot of different commodities and products are going to sell or stay in value – cars, homes, electronic appliances and consumer gadgets, clothing, luxuries, and the credit markets as well. We can learn something about the importance of such sentiments by looking back precisely at the problems that led to the adulation and coronation of VaR as the ultimate “quant” tool. We can see how important were PR factors in this growth curve and how it was bolstered and reinforced even when there were plenty of warning signs that the surface of things was not only descending but getting very slippery and icy. The following remark tells something important. As Guldimann wrote years later, “Many wondered what the bank was trying to accomplish by giving away ‘proprietary’ methodologies and lots of data, but not selling any products or services.” He continued, “It popularized a methodology and made it a market standard, and it enhanced the image of JPMorgan.” 8 Again, we are dealing with image, expectation, reputation, PR, image. The same phenomenon is how Madoff got away for so long with so much. Now if we can get at these images, reputations, outlooks, opinions, infatuations, faiths – both pro and con – using reasonable sentiment-uncovering tools (and other resources and know-hows), then we can have a handle on being able to inform: HERE is one reason X is catching on, being viewed as In, and Y is viewed as Out, being discarded. That gives information that is not explicit because it is not a number that anyone is tracking. Ironically, it seems that the LTCM and other 1998 upheavals (let’s not forget Russia as well, the grand collapse of ’98, and people like George Soros) did in fact trigger something that reinforced VaR while pointing the finger at “human psychology.” But at the same time people were not allowing themselves to see and accept the same problem in their own behavior. Why? Because of the rat race to grab as much as possible and to “not lose out” on what appeared to be a feeding frenzy for the “herd of caribou,” with no “predator wolves” anywhere near the horizon. Within most of the Wall St. firms, the pressures were on, not only to win big bonuses, to stand out as champions of profiteering, but to avoid the castigation and censure of management that held low regard for those who would sacrifice short-term gains for conservatism and stability over the long haul. This “human nature” issue is what clearly influenced the whole matter of what banks could use to regulate their deposits to cover shortfalls and losses. What investor would be comfortable with a company in their portfolio using its own internal VaR estimates to describe its position? And as we look at all the things that VaR did not do correctly, or at all, because of too short time (sampling) 8

Nocera [1], p. 7

8

Paradise Lost, and Risk Management Regained

mjdudziak

histories and simply not doing the measurements (e.g., liquidity risk), one thing really stands out, and it actually is complementary to the remarks that the fault is not in the mathematics but in how people use it. It’s all about what people will do to keep from hearing and seeing things that they do not want to hear and see, when they are in a situation where they think they have control and things are going their way or the way that they at least believe is going to be better for their egos.

Now let’s look at a couple of other points in this light. Take the remark, “It’s like the historic data only has rainstorms and then a tornado hits.” Consider the point about VaR being “gamed,” manipulated in such a way so that you get a lot of small gains consistently, even though there is this nagging 1% out there for a catastrophic loss. All of this goes back to the same thing about the human factor, and the three chimps above. “The issue, it seemed to me, was less what VaR did and did not do, but how you thought about it.” 9 But what does it mean to us and how can we use a better sentiment-finding tool to do what many would say is impossible: find the anomalies, and the sentiments, and the trends, and then transform that information from data into pointers about things that are either (a) the 1% or less that are known as possible, or (b) the 1% or less that we simply have never experienced before, period, at all? We must deal with (b) and acknowledge that there are a lot of (b) events out there. This is because, like it or not, we are in an incredibly first-of-its-kind world situation. Each of the following is a kind of phase-shift in its own right. • global world • huge population • energy shortages and irregularities • pandemic risk • climate change o global warming o nonlinear flip-flops and unexpectedly hot, dry, cold, or wet events • economic long-term depression 9

Nocera [1], p. 8

9

Paradise Lost, and Risk Management Regained

• •

mjdudziak

inseparability and non-insularity from influences out of all corners global conflict of an increasingly terroristic and asymmetrical nature

In short, we are living like someone who is in a filled-to-the-brim bathtub, there are a bunch of frayed and sparking electric wires laying on the floor and hanging overhead, you want to get out of the tub, the sides are oily and slippery, and a monkey is threatening to toss a big bean bag into the tub and create a certain disaster. Is this analogy too extreme? I think not, considering the scope of the falls from grace, and standing out not as a few billion in one or two quarters, but consistent multi-billion dollar setbacks by most of the prestigious banks worldwide. One thing we can do is to focus on another area that is somewhat like searching for anomalies, but it may be better to call this, “looking for the holes, as opposed to things that ‘stand out’.” Read this excerpt: That $50 million wasn’t just the most you could lose 99 percent of the time. It was the least you could lose 1 percent of the time. In the bubble, with easy profits being made and risk having been transformed into mathematical conceit, the real meaning of risk had been forgotten. Instead of scrutinizing VaR for signs of impending trouble, they took comfort in a number and doubled down, putting more money at risk in the expectation of bigger gains. ‘It has to do with the human condition,’ said one former risk manager. ‘People like to have one number they can believe in.’” 10 Now if we can identify the “holes” in institutions, systems, processes, and even people (experts) where things are not being examined, not being considered, looked-at, evaluated, where things are being letslide, then we can identify risks of a different and very critical and dangerous sort – the risk that an institution is not watching out for the icebergs, rocks and shoals where they are sailing their ship. Can we identify non-attention (inattentivity)? Can we pick out, thru reports, spreadsheets, dialogs, emails, blogs, other material, the absence of looking at something, the inattention to some sector or set of relationships? I believe that we can. We certainly can do so in other domains of activity, such as whether or not a security guard has done his rounds properly, or an inventory has been inspected, or if a driver is getting drowsy and not looking at some field of vision where he should be looking. Yes, we can do this if our tools can pull in large amounts of freeform, open-source, and varied, not preconditioned, data including text, not only tables of numbers! Nocera’s final paragraph says it very succinctly: “You just had to know that there were risks they didn’t sniff out — and be ever vigilant for the dragons.” So our task is to look for dragons, and particularly nearly-invisible ones with good camouflage. This is where the tools that I keep alluding to come into the foreground, and it is time to mention them, directly, because they are so appropriate to the tasks at hand. Looking for dragons in the corridors and canyons of Wall St. can be aided and abetted by looking at how the people up and down the streets are moving and what they are saying. In other words, making use of the sentiments in not only the media per se but in the “talk of the town” – the blogs and forums, the chit-chat and also the serious articles – can provide what is not yet turned into black and white numbers for production, sales, revenues. Sentiments tell about future courses of action, decisions, and positions. This is more in text and in the indirect discussions and expressions of opinion than in precise reports, but getting at this information means having a way to deal with very large masses and 10

Nocera [1], p. 9

10

Paradise Lost, and Risk Management Regained

mjdudziak

streams of text, and being able to perform computational linguistics that carries from the story to the mood. Patterns and Predictions is a toolset that uses Bayesian networks to make forecasts based upon probabilities and dependencies between those probabilities. As a software suite it is not an “out of the box” package but a set of modules and engines that include graphically oriented visualization tools for setting up relationships to be tested, pulling in the data from text sources including off the web, and directing output from the inference engine to other applications such as an organization may already have in place for management decision support. No one tool is the be all and end all of forecasting needs. However, what Patterns and Predictions offers is a kind of versatility that is both accurate and timely, and unlike many other statistical including neural network technologies, Bayesian networks are well suited for dealing with large volumes of textual data from non-specific sources – exactly what is needed to be analyzed in order to get at the sentiments and moods of large populations. One of the other things that we can do with Patterns and Predictions has to do with trigger-tests for data coming through the pipeline – for instance, using agents that run lots of simpler statistical and quick binary-decision tests on things “floating” in the stream, observations and comments connected to the user’s business topics of interest. This method offers yet another way to measure changes in how people are talking, and what is both standing-out and also conspicuous for its absence. Let me make an analogy of sorts. We are looking for well-camouflaged Ninjas and dragons lurking in the woods. Well, we need to pay attention to shadows. They may be hidden but they will cast shadows (no vampires considered here!). The shadows are where things are not being seen, looked at considered – this is about analysis of what analysts, brokers, traders, risk managers, are not talking about. There are also the strange anomalies. By this I mean the anomalies that matter because of the type, not only the degree. These are the rustling of leaves or the chirps of some birds because they, whom we cannot see or hear well, see or hear something (dragon, ninja) that is invisible to us. So we want to find not just any anomalies, but those that really “don’t fit or belong” in the setting. These are anomalies that can be of all sorts and types. Certainly in the mortgage and housing world there were plenty of such shadows having to do with prices, terms, rates, credit offerings, and all the way from individual transactions and properties all the way to packaged securities, and of course, those (in)famous credit default swaps. By employing textstream-oriented Bayesian analysis, there is an opportunity to have in hand precisely the kind of information that will put out strong red alerts as to when a “1%” outlier is coming on strong and when the standard model, such as VaR, needs to be complemented and supplanted. This does not mean that all the managers and decision makers are necessarily going to listen, rethink, and react, but it does mean that the signs will be much more evident, to larger numbers of people in the institution, so resistance to the wake-up call will be reduced. The information is, as almost always, out there in the world and accessible. The question is whether or not companies now will take advantage of the means at their disposal to hear and to read that information.

11