Having a model forecast is no guarantee of success. Memory might tell you the bar was crowded last weekend, so you decide to stay home this weekend because you assume the bar will be crowded again. This model says the future will resemble the past. If enough people have the same model and stay home, the bar will actually be uncrowded this weekend. Memory makes you lose out on an enjoyable evening of live music.
Conversely, the anticrowd remembers the bar was too crowded last weekend and decides people will go somewhere else next time. In their model, the future does not resemble the past. They decide to try the bar again next weekend. If they’re lucky, they’ll get a good table.
Still, if the anticrowd gets too big, the bar may end up overcrowded again. Then some anticrowd members may join the crowd and stay home. The bar might have empty tables next time, and so on.
On rare occasions, the random group might all choose the same course of action (like tossing five heads in a row) and cause crowd members to join the anticrowd, and vice versa, as adaptive behavior takes hold. This random behavior is a catalyst for shifts between crowd and anticrowd allegiance—a snowflake that starts an avalanche.
Scientists have conducted experiments using these same crowd-anticrowd dynamics. A group of individuals begin with a preference in their forecasting model. Through experience and feedback they self-organize into crowds, anticrowds, and random actors. The crowd and anticrowd attract the vast majority of participants in roughly equal numbers, while the random actors represent a small minority. This illustrates one of complexity’s most powerful traits: emergence. Well-defined opposing groups emerge without force or prearrangement from an undifferentiated mass through the workings of feedback and memory.
Emergent behavior is well documented in complexity science. It makes intuitive sense also. A Wall Street cliché says, “For every buyer, there’s a seller.” In a bull market, buyers are a crowd who believe the future resembles the past. Sellers are the anticrowd who believe that the future will be different. With both in equal proportions the markets can function. What about the small minority of coin tossers? Their individual behavior is random. Do they cause markets as a whole to be random? Or do they cause bulls to become bears, and vice versa, producing nonrandom persistence?
Research conducted by physicists Neil Johnson, Pak Ming Hui, and Paul Jefferies using financial market data shows the price movement pattern in markets does not correspond to the so-called random walk model that is the foundation of modern financial economics. Instead, behavior corresponds to predictions of complexity theorists using principles of feedback and adaptive behavior.
Behavior in financial markets can be broken down into binary choices, expressed as “either/or” or “yes/no” answers to a series of questions. Will you trade stocks today? Will you consider IBM shares? Will you buy or sell? Will you transact in large or small size? And so on. Each of these questions has a yes or no answer. In binary code, yes can be expressed as the digit 1. No can be expressed as the digit 0. Answers to a series of these questions can be expressed as a string of 1s and 0s such as 0011010011. These strings can be computer coded and analyzed for patterns in large data sets and long time series. The answers are highly revealing about how markets actually work.
The random walk model, associated with Princeton professor Burton G. Malkiel, says these decisions resemble a drunk walking down the street. Each step is uncertain. It could be forward or backward. The drunk doesn’t know himself. Each step is random, unaffected by the one before. There is no memory, there is no feedback.
The random walk model and the crowd-anticrowd model should produce completely different patterns of 1s and 0s because the random walk has no memory and the crowd does. Patterns produced by each model are quantified, and the model projections compared to experimental data.
Neil Johnson and other physicists do this by beginning with a thought experiment. Imagine the market as a person walking from a fixed point for a certain time period. The walker can move forward or backward the same way the market goes up or down. You want to calculate the distance covered. The purpose is to see if markets are a random walk or something else.
For convenience, scientists give the starting place a value of 10. Every step forward adds 1 to this position. Every step backward subtracts 1. From a starting place of 10, if you took two steps forward and one backward, your ending position is 11 (10 + 1 + 1 − 1 = 11). This forward/backward value is the same binary output as the yes/no investor choices described above and allows for binary coding and analysis.
This binary walking pattern from a starting position of 10 means that after 9 steps, the walker will either be at position 19 (10 + 9 = 19) or position 1 (10 − 9 = 1), or somewhere between 1 and 19 depending on the pattern of the walk.
For example, if you start at position 10 and take 9 steps forward you end up in position 19. From that starting position, the new position after each step presents this pattern: 10 11 12 13 14 15 16 17 18 19. This does not appear to be random; the path appears directional. Scientists call this pattern highly ordered.
To generalize this experiment for all types of walks, scientists create a measure that describes distance as a function of time. This function is expressed as ta where t equals the number of moves, a is an exponent, and ta equals the distance traveled. Both t and ta can be observed empirically in experiments. The exponent a is derived from results for t and ta.
In our ordered example, t = 9, the number of moves, and ta = 9 because that was the distance traveled. Therefore, a = 1 in this example; an exponent 1 applied to a number equals that number. In this highly ordered case, 9 = 91; the number of steps taken equals total distance traveled.
What happens when steps are truly random? In that case, the total distance traveled would rarely equal the total steps taken because some steps would be backward, some forward, and they would cancel each other out. The steps taken are greater than the distance traveled, which means that t > ta. If that is true, then a < 1, because a fractional exponent is required for t to be greater than ta.
For the random walker, many sequences are possible because there are numerous possible combinations of forward and backward movement when taking nine steps. Each step taken by the random walker is like a coin toss that can come up heads or tails. For analytic purposes, say that heads = 1, tails = 0, and each 1 is a step forward, and each 0 is a step back from a prior position.
As an experiment, I tossed a coin nine times and got the following sequence: 110001001, a total of 4 heads and 5 tails. Starting from position 10, and following the walk represented by these random coin tosses, the position sequence is: 11 12 11 10 9 10 9 8 9. In this random walk, the walker moved 1 position (10 – 9 = 1) in 9 steps. This random sequence is referred to by scientists as disordered because the sequence does not show strong persistence in one direction or the other.
If this experiment was repeated 1,000 times, easily done on a computer, the average distance from the starting place produced by the random walk of 9 steps would be approximately 3, which is the square root of 9. The distance of 3 = ta in our model. If t = 9 (the total steps taken), and ta ≈ 3 (total positions moved as shown by the random walk output), then a ≈ 0.5. The total movement in the 9-step random walk is 3 = 90.5.
In a highly ordered walk a = 1.0. In a random or disordered walk a = 0.5. What type of walk do actual markets take? Stated formally, what is the value of a based on actual market price movements?
One characteristic of complex systems is they are neither highly ordered nor random. Complex systems oscillate between order and disorder. This oscillation comes from agents’ deciding to quit the crowd and join the anticrowd or vice versa. A complex system that begins with random behavior can become ordered through feedback and adaptive behavior. Likewise, a highly ordered system can fall into disorder.
Complex systems move back and forth, exactly as markets move from bull to bear phases as investor sentiment moves from
fear to greed. Adaptation produces patterns more persistent than a random walk, tending toward order. Still, the system does not become completely ordered because of the crowd-anticrowd dynamic. In other words, the value of exponent a in a complex system should be somewhere between 0.5 and 1.0.
Empirical research on stock markets around the world over extended time periods shows the value of a in real markets is approximately 0.7. This empirical result falls between 0.5 and 1.0, somewhere between random and ordered, exactly what complexity theory predicts. This is strong evidence that capital markets are complex systems.
Not only do capital markets fit a descriptive definition of complex systems based on diversity, communication, interaction, and adaptive behavior, but empirical evidence shows that actual behavior in markets corresponds to the predictive output of a theoretical model. This is science at its best.
The implications of this conclusion are troubling. Professor Neil Johnson puts the matter starkly:
The standard model that most of the finance world uses to calculate how markets move is not accurate. . . . Financial markets are Complex Systems and they cannot be described accurately by anything other than a theory of Complex Systems. Standard finance theory may therefore appear to work for a while but it will eventually fail, for example in moments where strong movements appear in the market as a result of crowd behavior. And this is far from being a minor flaw since it is precisely these moments when your money is most at risk.
A grasp of complexity theory is a powerful tool for assessing risk in capital markets. We see how diverse groups of agents can self-organize into a crowd and anticrowd that produce a reasonably stable, yet not random, market. There is persistence, but not complete order.
Panic arises when the crowd and anti-crowd act as one. A completely ordered market system is one in which there are all sellers and no buyers. Such a market would instantaneously collapse and prices would go to zero. How likely is this? In natural complex systems, collapse happens with some frequency.
The New Madrid Seismic Zone in Missouri and surrounding states has been relatively stable for more than two hundred years. Yet in 1811–12, New Madrid produced four of the largest earthquakes in recorded North American history, estimated at magnitude 7.0 MW or greater (MW connotes moment magnitude scale, successor to the Richter scale). Seismologists predict the next New Madrid earthquake could be magnitude 7.7 MW, about the same as the 1906 San Francisco earthquake. These estimates include 86,000 casualties and two million people seeking shelter. Seismic faults are complex systems; so are stock markets.
The fact that complex systems oscillate between randomness and order does not mean these systems are stable or self-equilibrating. Complex systems go through phase transitions into chaos or collapse with surprising ease. The types of Bayesian simulations conducted at Los Alamos help the analyst to envision a range of outcomes in complex systems, including the financial equivalent of a thermonuclear bomb.
The financial outcomes considered in this book have happened before. Investors may lose money in a market crash, yet markets do tend to bounce back over time. Some crashes are excellent buying opportunities for those on the sidelines with cash who find bargains amid the rubble. Even those who suffer market losses can recoup if they hold their positions instead of selling into the panic. Most markets gain value over time. The lucky few who sell out at tops, and buy after crashes, outperform market averages.
Likewise, the failure of prominent banks seems to be a problem society has learned to manage. Stock investors in failed firms may suffer losses, yet depositors and account holders are routinely bailed out by deposit insurance and government guarantees. Even stock losses are manageable if they are part of larger diversified portfolios. After the crashes in 1987, 1998, 2000, and 2008, the market bounced back and rose to new highs. Why should investors be concerned about collapse?
The archetype for collapse in a complex system is not New Madrid or San Francisco. The archetype is Krakatoa. In 1883, the island of Krakatoa in the Sunda Strait between Sumatra and Java exploded with a force thirteen thousand times greater than the Hiroshima bomb. The force was ten times greater than the Castle Bravo bomb test on Bikini Atoll in 1954. It was four times greater than the largest nuclear explosion ever, the fifty-megaton Tsar Bomba test by the USSR in 1961.
After the 1883 Krakatoa explosion, there was nothing left of Krakatoa. The cause for investor concern is that certain systemic collapses are so large the system does not bounce back. The system ceases to exist.
CHAPTER 4
FORESHOCK: 1998
I have reflected a long time on the Long-Term Capital Management crisis. The thing that struck me most was the story of LTCM . . . is a very modern crisis, but the way it was resolved was almost identical to the way that crises always used to be resolved. The central bank was brought in and banged a few heads together. There was an argument about whether they should have done it, but in the end, that was how it was resolved.
Stanley Fischer, vice chairman of the Federal Reserve Board
God gave Noah the rainbow sign, no more water but fire next time.
“Oh Mary Don’t You Weep,” gospel song
Money Machine
The 2008 financial panic inspired a legion of books and movies, including a memorable narrative, Too Big to Fail, by Andrew Ross Sorkin. By all accounts, the financial system suffered a heart attack that year. The medical metaphor is not a stretch. The world financial system really did have a heart attack, and the patient nearly died. The Fed was a doctor with a defibrillator. What struck me most about 2008 was I had seen this movie before.
Exactly ten years earlier, almost to the day, the financial system suffered its first global heart attack. Fed doctors saved the patient that time too. Yet after 1998 the patient returned to cigars, heavy drinking, and no exercise. A second heart attack was just a matter of time.
If a market diagnostician in 1998 had studied the symptoms and course of that panic, the 2008 debacle could have been avoided. Nothing of the kind occurred. The lessons of 1998 were not learned. Dysfunctional market behavior resumed on an even larger scale with the blessing of banks and regulators.
The 1998 panic, catalyzed by Russian default and the collapse of hedge fund Long-Term Capital Management, seems small in hindsight. Many have not heard of it. Compared with the 2008 panic, the late summer of 1998 seems distant, inconsequential.
Superficially, the problems in 1998 seemed to go away. A few banks, notably UBS, took large write-downs. Some bank executives got fired. Alan Greenspan cut interest rates twice—once at a scheduled Federal Reserve meeting on September 29, 1998, and again at a rare unscheduled meeting on October 15, 1998. The second rate cut did the trick. It told markets the Fed was watching and would do whatever it took to restore calm.
Normality returned. Credit spreads that widened beyond belief began to converge. The stock market shrugged off the LTCM panic and resumed one of the greatest rallies in history. The Dow Jones Industrial Average soared from 7,632.53 on October 1, 1998, to 11,497.12 on December 31, 1999; a 51 percent gain in fifteen months. After LTCM left the headlines, it was as if nothing had happened.
Yet something did happen that had never happened before. Major stock and bond markets around the world were hours away from collapse. The biggest banks were set to fall like dominoes starting with the perennial weak link Lehman. Investors stood to lose more in relative terms than they did in 2008. This was not reported at the time despite intense media focus on LTCM and its reclusive founder, John Meriwether. Only a few insiders at LTCM, the Fed, Treasury, and foreign finance ministries saw the whole picture and understood its significance. Elites foamed the runways and brought LTCM in for a safe landing despite four engines in flames. Global investors were inside the plane with seat belts fastened and no exit. What seems in hindsight like a nonevent was a near miss of potentially catastrophic proportions.
The insiders who saved t
he system were quite prominent at the time. Some became more famous, or infamous, in later years. Peter Fisher led the emergency response by the Fed; he later became vice chairman of mega–wealth manager BlackRock. The bridge-playing Jimmy Cayne, head of Bear Stearns, was LTCM’s broker. He had the best information of any outsider on LTCM’s market risks. In typical Wall Street style, Cayne refused to share the information with other bank CEOs and nearly wrecked the rescue.
Jon Corzine, head of Goldman Sachs, was a leader in the LTCM rescue along with the CEOs of Citi, JPMorgan, and Merrill Lynch. Corzine, a close friend of Meriwether’s, was distracted during the bailout because his own trades were losing billions. Corzine’s debacle at MF Global in 2011, ending in bankruptcy, came as no surprise to those who watched his reckless gambles at Goldman.
At the height of the 1998 panic, Goldman pirated LTCM’s derivatives positions and used the information to cover their own trades and front-run competitors. Goldman also tried to wreck the LTCM bailout deal at the eleventh hour by front-running the Federal Reserve with a competing offer signed by Corzine, Warren Buffett, and AIG head Hank Greenberg. AIG met its own demise in a government bailout in 2008. Cayne and Corzine were among the heads of the “fourteen families,” the fourteen Wall Street banks that participated in the LTCM bailout.
The LTCM rescue would not have succeeded without discreet intervention by legendary bank crisis maestro Bill Rhodes, who finessed foreign bankers and finance ministers for loan waivers while the fourteen families squabbled among themselves. Unknown to outsiders, LTCM had arranged almost $1 billion of credit on an unsecured basis from a nineteen-bank international syndicate. To complete the rescue, waivers were needed from those banks in addition to getting the fourteen families to infuse new cash. Rhodes got those waivers.
The Road to Ruin Page 13