CHAPTER 5
Biological Frameworks: EPIDEMIOLOGY AND EMERGENCE
Even completely rational people can participate in herd behavior when they take into account the judgment of others, and even if they know that everyone else is behaving in herdlike manner. The behavior, although individually rational, produces group behavior that is, in a well-defined sense, irrational.
—Robert Shiller
This chapter provides two biological lenses through which to study booms and busts: an epidemic lens and an emergence lens. The epidemic lens, as described next, has use in helping us to determine the relative maturity of a boom and the potential imminence of a bust. The emergence lens provides a powerful explanatory framework through which to understand how groups can be misled into an uninformed consensus.
Scientists and medical professionals alike have been studying the dynamics of epidemics for hundreds of years. The basic framework utilized by these practitioners has been one focusing on infection rates. Many variables complement this focus, but if we recognize human behavior typical of a boom as “feverish,” then the analogy becomes more obvious. Although the “infection rate” (for example, how quickly people believe the world is different) is not very useful by itself, combining it with the rate at which people are either “cured” of their disease or die from it exponentially increases its value to us. The chapter briefly touches on these dynamics before turning to the idea of emergence.
Emergence is the study of seemingly chaotic efforts by large groups of animals that tends to produce extremely robust and adaptive order. Herds and swarms are among the most prominent of such phenomena in the biological arena, but examples exist in urban planning and other domains. This chapter briefly discusses herd and swarm behavior in social insects and animals before evaluating these dynamics in humans. We close the chapter by considering the ramifications of these tendencies for financial markets.
Revealing the Maturity of an Unsustainable Boom
Epidemiology is the study of diseases and their transmission across a population. If we think about markets as being composed of individuals who are either affected or not affected by a particular “disease” (i.e. infatuation with a new thing), the basic terminology of epidemics has a striking pertinence to the study of booms and busts.
Epidemiologists have developed extraordinarily complex and intricate models of disease transmission. The most basic elements of all these models, however, are the infection rate and the removal rate. The infection rate is the rate at which the disease is transmitted from infected individuals to those who are susceptible to infection. The removal rate is the rate at which infected people are removed from the population of transmitters, either because they die or because they have recovered and are now immune to the disease. Although relapses may prove possible, for purposes of our discussion, we assume that those who contract the disease either die or recover into an immune state.
To better understand how these two rates interact, let us consider three primary scenarios: (i) infection rate > removal rate = 0, (ii) infection rate > removal rate > 0, and (iii) removal rate > infection rate > 0.
Our first scenario, in which the infection rate is greater than the removal rate and the removal rate is 0%, produces an epidemic that eventually infects 100% of the population. The pace at which the disease spreads will depend on the infection rate. Graphically, any scenario that involves a removal rate of 0% will follow a logistic curve, as in Figure 5.1. As can be seen in the graph, the initial percentage of the population that is infected rises slowly at first. Because the removal rate is set to 0%, the population is infected at the infection rate. As summarized by Robert Shiller in Irrational Exuberance:
Figure 5.1 Infection Rate > Removal Rate = 0%
Although the rate is nearly constant at first, the absolute number of people recorded as contracting the disease rises faster and faster: as more and more people become contagious, more and more people become infected … but the rate of increase starts to decline as the pool of yet-to-be-infected susceptible individuals begins to be depleted.1
Eventually, 100% of the population is infected, as seen by the flat line at upper right in the graph.
The second scenario is one in which the infection rate is greater than the removal rate, but the removal rate is greater than zero. In this case, the cumulative distribution of the infected population will resemble a bell curve. The graphical depiction of the infected population will rise from zero slowly and then accelerate before peaking and returning to zero. Although 100% of the population might still get infected in reality due to random factors, it is highly likely the peak will occur before 100% of the population is affected. Figure 5.2 summarizes this scenario.
Figure 5.2 Infection Rate > Removal Rate > 0%
The third scenario in which the removal rate is greater than the infection rate which is greater than 0% is particularly uninteresting as it provides for no epidemic. People are cured more quickly than the disease is spread, resulting in no cumulative infection of the population. The best analogy for this situation is a noncontagious disease.
Although the formal academic application of epidemic models to financial markets has not gained significant traction, there has been some research conducted on word-of-mouth dissemination of ideas. Ideas, however, are not nearly as consistent as biological diseases and are subject to transmission errors via what Shiller calls a “mutation rate.”2 Shiller goes on to note that the transmission of ideas is similar to the children's game of telephone in which an original idea is so distorted via transmission as to be laughable after several transmissions. Further, alternative newsworthy items (such as a murder, car crash, or death of a major celebrity) may in fact affect the infection rate in a manner that deflects attention from the idea being transmitted. Thus, the relative prominence of an idea and the attention it therefore receives will affect its transmission rate.
Despite these concerns, the overarching objective of presenting this lens is to provide a vocabulary for thinking about the spread of ideas through a population during speculative bubbles. The framework of infection, removal, and mutation rates provides a useful conceptual method of understanding. Although Shiller is clear in describing the limitations of applying an epidemic model to idea diffusion through a population, he highlights the usefulness of the framework for thinking about speculative manias, emphasizing the need for a transmittable story:
Word of mouth may function to amplify public reaction to news events or to media accounts of such events. It is still necessary to consider the infection rate relative to the removal rate in order to understand the public impact of any new idea or concept, since most people's awareness of any of these is still socially mediated. Thus the likelihood of any event affecting market prices is enhanced if there is a good, vivid, tellable story about the event … . Word-of-mouth communications, either positive or negative, are an essential part of the propagation of speculative bubbles.3
Ultimately, the usefulness of an epidemic lens is its ability to help gauge the approximate maturity of an unsustainable boom. Just as a bubble that is relatively more progressed (i.e. closer to the point of bursting) will have more of its population infected, so too will an earlier-stage bubble have a large “yet-to-be-infected” and susceptible population. Because it is very difficult to gauge with any precision the percentage of the population that is infectable or that has been infected, red flags or indicators that reveal approximations are useful. For example, a growth in amateur or beginner investors into a particular asset class is very telling. After all, we must assume that expert investors have already been infected, and if amateur investors are now infected, who is left to infect? Thus, by gauging the prevalence of newcomers to an asset party, we can get an approximate sense of the bust's imminence.
How Micro Simplicity Drives Macro Complexity
In many biological examples, group actions that appear ordered and deliberate emanate from uncoordinated individuals. How is it that these collections of social animals appear to operate in a seamless manner typical of a single organism? The study of complex biological systems has been the focus of significant research and has spawned a niche industry of “emergence” scholars. Despite being a relatively new area of rigorous research, early studies on the topic were taking place in the 1970s, albeit without the label of emergence. One of the first studies on this topic involved the behavior of Atlantic pollock.4 University of Miami biologist Brian Partridge took it upon himself to gather schools of 20–30 fish (each about 3 feet long and weighing 40–50 pounds) and have them swim in a circular tank (33 feet in diameter) while he observed them from above, while spinning himself in a manner that kept him geospatially above the moving fish. Each fish was labeled in a manner that allowed him to track individual behavior.
The movement of each individual fish was then analyzed by reviewing more than 10,000 frames of film and Partridge's real-time observations. After completing such painstaking research, Partridge and his research team concluded that individual pollock followed two simple rules to move as if they were a single unit: it was as if they were being told to “swim behind the fish directly in front of you” and “swim at a speed that keeps pace with the fish next to you.” As simplistic as these two rules seem, they appear adequate to explain how the school manages to react when threatened by a predator or seeking to avoid an obstacle in the water.
During the summer of 2010, I decided (don't ask me why, because I'm not sure I know the answer!) to compete in an international distance triathlon. In an effort to provide a “unique experience” to the athletes, the race director decided to send hundreds of swimmers into the ocean at once. Over the course of the roughly 1-mile swim, I was often stuck in a group of swimmers. If I sped up, I found I ran into other athletes. If I slowed down, others ran into me. When I tried to move to the left, I hit another swimmer. The same thing occurred when I tried to move right. Eventually, I decided to settle into following the rules that Partridge and his team found drove the fish in a school … and I was shocked by the overwhelming sense of peace that overcame my efforts when I did. Not surprisingly, several observing friends noted that “it looked like a school of fish were swimming together… . The whole group of you managed to turn and zigzag for no apparent reason, but almost as if someone were directing you all to do so.” Knowing full well that none of us were coordinating anything, I was impressed by the sense of group order that came out of what was utter chaos at the individual level.
As it turns out, swarms are examples of emergent complexity that originates from the following of simple rules on the part of its members. No group consciousness that forces the subservience of individual action is required. As an example of how simple rules exercised by individuals might create group behavior that appears complex to external observers, consider the popular human wave at a ballgame. As noted in Dr. Len Fisher's 2009 book The Perfect Swarm: The Science of Complexity in Everyday Life, “The wave might look to a visiting Martian like a complicated exercise in logistics, but its dynamic pattern emerges from a simple rule: Stand up and put your hands in the air (and put them down again) as soon as you see your neighbor doing it.”5
To better understand how such emergent order occurs, we turn to the behavior of several social insects6 —locusts, bees, and ants—that have demonstrated the ability to generate coherent, organized, methodical group behavior, despite the seeming improbability of doing so with billions of individual members. After discussing how group complexity emerges from individual simplicity in insects, we consider the manifestation of this phenomenon in human behavior. Let's begin by understanding locust behavior.
Locusts
Of the more than 12,000 known species of grasshoppers, fewer than 20 are classified as locusts.7 Nevertheless, locust plagues—written about since biblical times—continue to affect more than 10% of the world's population.8 For this reason (as well as scientific curiosity), scientists have been very interested in understanding their behavior.
Locusts are different from other grasshoppers in one major and very important way: their behavior changes radically when they are placed into crowded situations. Grasshoppers tend to disperse if placed in close proximity to each other, but locusts tend to synchronize their movements. The conversion from chaotic crowd to orchestrated swarm occurs as the locusts find themselves in dense quarters. Research has shown that young locusts will move chaotically until the density of their crowd approaches seven locusts per square foot, at which point they begin marching in sync with each other.9
As it turns out, most marching is a quest for food, and given the extraordinary volumes of food—daily intake equivalent to their body weight—that locusts consume,10 cannibalism is not uncommon.11 This desire not to be eaten provides the motivation to keep moving. Although this might have seemed obvious to our pollock-watching biologist, it is a useful third rule to explicitly articulate: “Avoid hitting the member directly in front of you or being hit by the member behind you.” Given this avoidance desire, why don't the locusts simply disperse? The best way to avoid being eaten is to avoid your fellow hungry locusts, right?
As noted by Len Fisher,
Normally shy and solitary, the close proximity of other locusts … stimulates them to produce the neuro-modulator serotonin, which not only makes them gregarious, but also stimulates other nearby locusts to generate serotonin as well. The ensuing chain reaction soon has all the locusts in the vicinity seeking each other's company.12
Literal party animals! A physiological response then creates more mobile locusts that begin moving in swarms, initially on the ground and then in the air. Just as the serotonin, which is released in escalating amounts as the serotonin in close-by locusts rises,13 drives the desire to be with each other, so too does such intense partying make the locusts hungry—thereby assuring some healthy distance between each of them. As the party progresses, it gathers more members, until dense swarms of around one hundred billion (100,000,000,000!) locusts cover areas of up to 500 square miles.14
Although locusts are helpful in demonstrating how group behavior emerges from seeming chaos through the individual member's application of three simple rules (avoidance, alignment, and attraction), they fail to show us how a swarm makes decisions and develops a group logic that is different from an individual's logic. To see how decision-making in swarms takes place, let us now turn to bees.
Bees
Bees are social insects that tend to follow the three simple swarm rules as they travel in groups. Where they differ, however, is in the group's ability to head directly for a particular target (for a new hive, or a food source) identified by scout bees. How is it that a group of uninformed bees in a swarm are able to efficiently find their way to a target? Although much has been written about the famous “waggle dance”15 that scout bees conduct in their hives to communicate the direction and distance to a target,16 recent evidence suggests this communication is not sufficient to explain the swarm's behavior. Fisher notes:
The dance is performed in a hive that is almost as dark as some discos, so only those bees nearby (about 5 percent of the total) see the dance. The majority doesn't see it, so most bees begin flying in complete ignorance. Those that have seen the dance aren't even out in front, showing the others the way. They are in the middle of the swarm, flying with the rest.17
Given the lack of obvious leadership in the swarm, efficient direction of the group toward a target seems a highly unlikely outcome. To better understand what might be going on, scientists used cameras to capture individual bee movement behavior to see if they might identify some possible explanation for the group's flight pattern.18 By photographing the bees from below and leaving the camera's aperture open for a short time, researchers were able to produce a “map” of a bee swarm in motion in which each bee's movement was a short line. Most of the lines were short and curved, but a handful of the tracks were a bit longer (indicating greater relative speed) and pointed straight at the target. Those speedy bees heading straight for the target were labeled “streakers” by the scientists and provide the answer to our question.
It turns out that if a group is following the three simple rules of swarm behavior (avoidance, alignment, and attraction), these informed bees are able to take an unsuspecting group of ignorant bees rapidly and efficiently to their target. By moving a bit faster than the group, the informed bees exert a silent leadership that the uninformed bees follow.19
Fisher eloquently summarizes this finding:
In other words, it needs only a few anonymous individuals who have a definite goal in mind, and definite knowledge of how to reach it, for the rest of the group to follow them to that goal, unaware that they are following. The only requirements are that the other individuals have a conscious or unconscious desire to stay with the group and that they do not have conflicting goals.20
Ants
This methodology of the few informed animals leading a group directly toward a target is not the only form of emergent swarm intelligence that has been observed in social insects. Ants have proven to be equally effective at finding direct routes to food sources and other targets, yet their methodology is entirely different from that pursued by the bees.
Research conducted on a colony of Argentine ants at the University of Brussels specifically sought to understand why it was that ants were able to efficiently and directly travel to their targets.21 Scientists working in the Department of Behavioral Ecology created a forked path between the ant colony and a source of food. One path was approximately twice as long as the other path. They found that although the initial ants chose randomly, within a few minutes the whole colony was utilizing the shorter route. How and why did that happen?
The answer is actually quite simple, and once understood, the efficiency of ants makes a great deal of sense. Ants emit a chemical substance called a pheromone, which attracts other ants. More pheromone attracts more ants more intensely, and pheromone dissipates with time. Thus, the ant that took the short path ended up getting back more quickly and producing an option for other ants that has at least twice22 the pheromone of the longer path. Because of the greater pheromone on that trail, the next ant is highly likely to choose the shorter path. Even ants that take the longer trail to the food will be more likely, again due to the higher pheromone levels, to return via the shorter route and to further increase the pheromone levels on that route. Within minutes, the pheromone levels on the shorter trail overwhelm those on the longer trail to the point that the entire colony begins using the shorter path to the food. Here again, we find that individual adherence to basic rules lies at the root of this collective behavior: “The colony's efficient behavior emerges from the collective activity of individuals following two very basic rules: lay pheromone and follow the trails of others.”23
The other interesting finding from the study of ants is that emergent behavior with silent leadership (even if random or unintended) was able to efficiently guide the swarm of ants to a target. Thus, the idea of the queen bee or queen ant has been refuted by research in favor of group-driven behavior, with significant implications for group decision-making in animal groups.24
Emergent Behavior in Human Swarms
We now turn to the study of a particularly social and group-oriented animal, the human being. Historically, animal decision-making processes have been studied from a “simple rules” perspective, whereas human decision-making processes have been focused on complex utility functions. Unfortunately, the two disciplines have not cross-pollinated their research efforts until very recently. The next section provides highlights emerging from this multidisciplinary research.
Swarm Processes and Group Dynamics
As we transition from social insects to humans, swarm logic is more robust and transferable than most might believe. Although the case studies later in the book will illustrate how swarm logic among investors might exacerbate booms and busts, this section of the chapter will briefly touch on the research that has been conducted on humans as social animals and in which group decision-making processes are the focus. Unlike Chapter 3, which focused on how individuals make decisions, this section will focus on how groups make decisions and how individuals within them affect and are affected by group dynamics.
We begin by reviewing some recent research conducted by biologists, zoologists, and ecologists, and mostly published in journals not typically read by economists, financiers, or social scientists. Several of these studies were conducted to determine if humans act similarly to social insects in their approach to making group decisions. The Philosophical Transactions of the Royal Society of London dedicated an entire issue to the topic of “Group Decision Making in Humans and Animals.”25 The findings, though not entirely shocking, have profound implications for our study of booms and busts.
One particularly interesting piece of research26 published as part of this collection of papers was about an experiment conducted on groups of college volunteers. Several students were told to walk anywhere in the room, as long as they stayed within one arm's length of another student, which effectively created both the avoidance and the attraction criteria needed for swarm conditions. Communication was prohibited, but around the room were several targets in the form of uniformly placed letters. Prior to the experiments, one or two students in the group were given secret instructions to head toward one of the targets. By the time the experiment was stopped, most groups had ended up at the target letters given to the informed students. The unsuspecting students had been “led” there by a small minority of focused and informed leaders.
Although it is interesting to learn that uninformed human groups can be led by covert but informed leaders, the world is often filled with multiple leaders frequently targeting different objectives. Because of this fact, researchers began investigating the behavior of uninformed groups in situations including multiple leaders with conflicting targets. An article in the scientific journal Animal Behavior summarizes the research (which is based on a similar research design to the one just mentioned, but utilizing two informed students who were given differing objectives): “When conflicting directional information was given to different group members, the time taken to reach the target was not significantly increased; suggesting that consensus decision making in conflict situations is possible, and highly efficient.”27
Basically, the same experiments were conducted with two individuals given differing objectives. The groups were no slower in reaching the targets and quite rapidly made decisions about the appropriate course to be pursued. The implications of this finding on the study of booms and busts are enormous. If unsuspecting group members (think ordinary investors) can be led in any direction by a relatively small number of confident (regardless of whether such confidence is merited) members, we can imagine how such confidence might feed upon itself to generate a boom-like scenario.
Famous research conducted by Stanley Milgram28 in the late 1960s demonstrated the power of silent leadership in groups.29 Professor Milgram arranged to have people on the street stop and stare up at a window on the sixth floor of a tall building. Len Fisher eloquently summarizes the findings of the research: “With just one person staring up, 40 percent of passers-by stopped to stare with them. With two people, the proportion rose to 60 percent, and with five it was up to 90 percent.”30
If nothing else, the research confirms the ability of a few people to guide the behavior of a much larger mass. Note that this experiment actually demonstrated the ability of a few to mislead the many, an outcome that has significant pertinence to the study of booms and busts.
Intentional misleading may not even be a motivation of the group's pioneers, but the idea that initial decisions matter is significant. Consider the following informal experiment I ran after having read a great deal about bee logic and the ability of silent leaders to generate consensus. After landing at the Las Vegas airport (an airport with which I have a bit too much familiarity), I was the first person to exit the plane. I decided to get off the plane and head right (the opposite direction from the main terminal), walking briskly and with the conviction of a knowledgeable passenger. Despite signs pointing in the opposite direction, I was amused to see that the next 10–15 passengers who got off the plane immediately turned right and followed the “crowd.” I repeated this experiment on eight other flights over the next six months in various airports, all with similar results. It definitely seemed that bee logic applied to humans! The key lesson of this ad hoc research was that humans have a tendency to conform to the behavior of the seemingly knowledgeable group member.
The next section demonstrates how initial decisions made without meaningful reason might be interpreted by later deciders to have been based on careful analysis and contemplation. By placing greater weight on early decisions made during a chain of decisions, information cascades may develop and create outcomes that seem to defy explanation.
Information Cascades and Herd Behavior
Imagine, for example, that two equally good restaurants open on the same street adjacent to each other.31 They are very similar in every way, including cuisine, price, and ambience. Suppose now a young couple comes by and must decide on which restaurant to choose. Given that both restaurants are empty at this point, they have very limited information with which to make a choice. After perusing both menus, they flip a coin. Heads they go to restaurant A, tails they go to Restaurant B. They flip the coin and end up in Restaurant A. The second couple that comes by now has the same information as the first, as well as the fact that it appears the first couple chose Restaurant A. Unable to decide based on the restaurants’ menus and atmospheres, they opt for Restaurant A, assuming the first couple must have chosen for a good reason. Likewise, a family of five comes by, and seeing Restaurant A fuller than Restaurant B, they choose Restaurant A. This process might continue until Restaurant A is full, while Restaurant B remains empty.
As stated at the outset, the two restaurants are virtually identical, so how is it that one received all the business, and the other remained empty the whole night? Given that customers made their decisions based on the decisions of those prior to them, an information cascade took place through which information—relevant or not, but embedded in the choices made by others—influenced the actions of those who followed.
It is believed that such information cascades serve as the basis of herd behavior and have deep evolutionary value. If the leading buffalo in a herd suddenly stops and moves right, it may be because it has seen a lion. If 50 buffaloes do that, it would likely be unwise for the 51st buffalo to dismiss this data. When it comes to the rationality of markets, however, such herd behavior can be quite distortive of market prices.
The Blind Leading the Blind
The biological lenses presented in this chapter have broad applicability to the study of booms and busts. Although the epidemic framework is a useful tool for evaluating the relative maturity of a boom and the corresponding imminence of a bust, it is only valuable in providing an approximate sense of timing. By no means can it generate the precision needed by active risk managers. It yields virtually no insight on a day-to-day basis and is unlikely to prove useful in timing the bursting of a bubble with precision. Nevertheless, thinking about financial euphoria as a disease that has the potential to spread through an entire population proves quite useful in gauging the relative maturity of a boom. The most telling signs of a mature boom that is rapidly approaching the bust phase are a rapid growth in the number and type of participants, as well as the increasingly prevalent participation of unsophisticated or amateur investors.
The implications of the emergence phenomenon for our study of booms and busts are quite dramatic. The applicability of information cascades and the restaurant example discussed earlier are easily understood: seemingly irrelevant decisions take on greater meaning than originally anticipated and have the potential to snowball into herdlike behavior of uninformed individuals. If everybody else is making money investing in housing, why shouldn't I? Clearly they've done the analysis and everyone can't be wrong, can they? Yet everyone relying on the fact that everyone else has “done their homework” can create a dynamic in which random decisions made earlier in the chain acquire unwarranted and unintended significance.
The jump to silent leadership from information cascades is not a particularly large leap. The connection occurs via the (seemingly) informed individuals, and just as actually informed individuals like our streaker bees can accurately lead a group of uninformed individuals, so too can silent leadership by seemingly informed (but actually uninformed) individuals lead groups astray. Recall my ad hoc experiment at the Las Vegas airport. Not only did my seemingly informed status (likely conveyed by the definitiveness of my direction and focus of my efforts) lead others astray, but this silent leadership snowballed in a small cascade. Like ant pheromone, each additional person who followed me provided “guidance” to exiting passengers to follow the crowd. Surely not everyone would be walking in the wrong direction, would they?
The fact that many uninformed individuals can be so easily misled by other uninformed (but acting as if informed) individuals is a powerful finding. Suddenly, irrational group behavior is more understandable. If random events (such as the selection of a restaurant) have the potential to snowball into information cascades that create (without reason) completely lopsided outcomes, the implications become quite clear and important: Efficiency and stability can easily be replaced by positive feedback dynamics that drive instability and tipping away (rather than toward) an equilibrium point.
This chapter concludes Part I of the book and the presentation of the five lenses that we will use in our case studies that begin in Part II. Let us now turn to Tulipomania and the first demonstration of the lenses in action.