Mauboussin's Use of Complex Adaptive Systems
Michael Mauboussin is the chief investment strategy at CSFB. As an investment strategist, Mauboussin's approach begins with the premise that the value of a stock, and hence, stock price, is the expected net present value of a future stream of free cash flows. Mauboussin stresses that the true economic drivers of value are free cash flow (cash-in versus cash-out), risk (and appropriate demanded return), and the Competitive Advantage Period (CAP; the sustainability of those excess returns). Thus, in his model competitive advantage is quantitative; it exists when a business's strategy organizes its tactical activities to secure returns in excess of its cost of capital.
What is new in Mauboussin's approach is his unique and valuable use of (a) Alfred Rappaport's views of "Cash Economics" (Creating Shareholder Value, 1986; 1998rev) to analyze the Income, Balance, and Cash Flow financial statements, and (b) the Santa Fe Institute's "Complex Adaptive System" (CAS) to understand the stock market.
Rappaport encapsulated his perspective in the pithy, "cash is a fact, profit is an opinion." Critical of a 500-year-old accounting system that tracks tangible assets, but not intangibles, he argued that "following the cash flow" is how to get a business to show you the money. Calling his reading of Rappaport's book an epiphany, Mauboussin developed and implemented his cash-economic-ideas into his own research and investment analyses, including incisive criticism of using rules of thumb, like EPS and the P/E ratio, as if they were drivers of value when they are not. Together, they are currently collaborating on a book.
For several years, Mauboussin has been involved in the work of the Santa Fe Institute, which is a think-tank for developing a new science of complexity to understand such non-linear dynamic systems as the stock market, the origin of life, and Earth's ecology. This fall, Mauboussin will host his second conference there on the new economy.
Complex Adaptive Systems
I have tried to collapse the major features of a CAS into a provisional definition that includes its descriptors, crucial properties, and principal defining features. I have drawn from Mauboussin's, Holland's, and my own, albeit limited, understanding of CASs. Examples are used to illustrate, explicate, and extend the meaning of the defining characteristics.
Definition. A "complex adaptive system" is defined as an open, life-like, emergent, self-organized, global whole, as a meta-system that orders phenomena in dynamic, non-linear domains. These domains include cells, embryos, immune systems, brains, ant colonies, ecologies, scientific communities, political parties, and the economy.
Consider the descriptors of a CAS: "open, life-like, emergent, self-organized, global whole." In contrast to a closed system, like the traditional economic view of an equilibrium system created by the laws of supply and demand, the CAS is "open" to new inputs from distributed agents that sometimes produce continuous change and sometimes discontinuously reorganize the meta-system. As an open system, one that receives input from outside the system, the CAS begins, at a certain critical point, to exhibit life-like behavior, metaphorically taking on a mind of its own. This is the moment when the hive swarms, traffic jams form, and birds flock without following a leader.
This life-like behavior is described as "emergent" when it operates at a new level of complexity, as if it were an autonomous agent itself. Emergence occurs when the CAS, as a meta-system, displays a meta-level of self-organized behavior that is different from the self-organized behavior of the agents in it. In the CAS, "self" is the global whole, the meta-system itself regarded as an "entity." (Perhaps, the phrase, "system-organized" might be intuitively clearer than "self-organized.") The self-organized whole displays critical levels where change becomes non-linear, more than the sum of its parts, becoming a resultant of multiplicative changes, the resulting product of its many interacting agents (members or elements). These metal-level processes develop over time, evolving from the interactions of its many distributed members. Each agent responds to his or her own feedback loops, learning from errors in responding to continuous changes or adapting to discontinuous changes in the environment. When positive or negative feedback loops in multiple agents reach critical mass, the CAS emerges.
Crucial Properties. According to John Holland, complex adaptive systems share some crucial properties. First, the CAS in each domain is composed of many living agents acting in parallel. In the brain, the agents are neurons; in the ecology, the agents are species; and in the economy the agents are individuals or households. Each agent exists in a fluid environment of perpetual novelty that is created by it multiple, changing interactions with other agents who are in multiple, changing interactions with still other agents.
Second, the control of the CAS is highly dispersed, not coming from a master control center in the brain or the cell or the economy, but arising from the dense web of competitive and cooperative interactions of the distributed agents themselves. A CAS has many levels of organization, with agents at one level serving as the building blocks for the agents at the next higher level. For example, proteins, lipids, and nucleic acids are the building blocks for a cell, cells that are the building blocks for tissue, tissues for organs, organs for the whole organism, organisms for an ecosystem, and so on. An important property of the CAS is its ability to adapt by constantly revising and recombining its building blocks as it gains experience in its environment. At a fundamental level, Holland believes learning, evolution, and adaptation are similar: each involve the combination and recombination of building blocks.
Third, all complex adaptive systems are anticipatory mechanisms. This is easy to see in an economy, for instance, where anticipated inflation influences consumer's to buy before the prices increase, which increases inflationary evidence that roils the stock market by depressing stock prices. But, Holland extends this idea beyond human consciousness when he notes that implicit conditions are encoded in the genes, making a prediction that this organism can do well in this environment. Through learning, the agent develops conditional rules, for instance, given circumstances A, B, C, that actions X, Y, Z will pay off. More generally, Holland posits that each CAS is making predictions based on its implicit or explicit assumptions about of how thing are out there in the world. He calls these assumptions "internal models." Like subroutines in a computer program, these internal models (adaptive schemata) can execute, producing behavior. They are the building blocks of behavior, capable of being tested, refined by feedback, and rebuilt by learning from experience.
Finally, the CAS includes many niches, each of which can be filled by an agent who adapts to the niche in competition with other agents. The filling of one niche, however, opens other niches; so new opportunities are constantly created. Equilibrium is never achieved because the CAS is constantly adapting to unending, ongoing change. Never optimizing their fitness to a set ideal, the most any agent can do is continually to increase relative fitness. Thus, complex adaptive systems are subject to perpetual novelty.
Defining Characteristics of Complex Adaptive Systems: A Complex Adaptive System: (a) emerges as a meta-system from the bottom-up, aggregated interactions of many distributed agents who have learned and are following local decisions rules (organized as schema), creating the appearance in the emergent meta-system of top-down, global decision rules; (b) demonstrates its emergent properties by acting-as-if it had life-like "agency," becoming something more than the sum of its parts, a something more that is produced by the multiplicative product of its interacting local agents, an emergence that can be described as the phase transition of "complexity," on the edge between order and chaos; (c) responds to trends in the domain toward increasing or diminishing returns that have developed from feedback loops in its distributed agents, feedback that has amplified (positive) or attenuated (negative) ongoing trends toward either increasing or decreasing returns, respectively, among its agents; (d) achieves an emergent system-adaptive-fitness from the competitive winnowing of local, learned schemata through an evolutionary-like process of selection and survival of the combined and recombined information in local descendents; and (e) achieves a higher-order complexity by demonstrating a self-organized criticality that is endogenous to the meta-system, exhibiting unpredictable and unprecedented nonlinear responses at critical points to changes in the domain.
A. Emergent from Aggregated Interactions. Craig Reynolds created a computer simulation of the flocking behavior of birds, the schooling of fish, or the herding of zebras and wildebeests. Reynolds placed a multitude of bird-like agents ("boids") into a sky filled with walls and other obstacles. Each boid was programmed to follow three simple rules for its behavior: (1) Maintain a minimum distance from other objects, including boids; (2) Match the velocity of boids in my neighborhood; and (3) Move to the center of the mass of boids in my neighborhood. Notice that this schema contained no rule that said, "Form a flock." No matter how the birds were placed before they began to fly, they soon formed flocks that flew over walls and around obstacles, sometimes splitting into sub flocks to fly around poles before resuming as a single flock. Thus, flocking is an aggregated emergent outcome that arises from many distributed agents following three local rules. A fluid and emergent order arose from complex interactions of agents following their own rules.
B. Emergence as Agency Deriving from Nonlinear Complexity. Mauboussin cites research from Carnegie-Mellon University that demonstrated that the market itself (another CAS)) substitutes for irrational investment decisions of individuals. "Traders" were given poor or nonsensical trading rules, but the simulated market remained remarkably efficient. Given the structure of double auction markets, even dumb agents generate smart results. Gode and Sunder concluded, "Adam Smith's invisible hand may be more powerful than some may have thought; it can generate aggregate rationality not only from individual rationality but also from individual irrationality." Thus, the stock market, like the Internet, does not require leaders to coordinate it (Do you hear that Alan Greenspan?), and its structure, not the individuals who comprise it, allows it to mimic the economic model. Markets, like flocks, have no leaders, no lead steers directing its bullish path; but the market itself may provide an aggregated basis for forming sounder decision rules for investors.
C. Feedback and Increasing Returns. A feedback system can be designed either to match a set standard or a changing one. When using a set standard, feedback from changed responses provides information about the degree of match between the present condition and the standard until the difference between the present level and the standard is reduced to zero. When feedback uses a ratcheting standard, that is, one changing iteratively, then the output of the first iteration becomes the input of the next iteration; in a continuing loop, this positive or negative iterative feedback produces an increasing or a decreasing series of returns. While he was reading the Belgian physicist Prigogine, the Irish-American economist Brian Arthur had a sudden insight: he recognized that the economy was not an equilibrium system that was regulated by decreasing returns back to equilibrium but was instead a self-organizing system that contained positive feedback loops. Instead of always dying away to equilibrium, a tendency, arising even from happenstance, could become amplified by positive feedback into a trend that might persist as a virtuous cycle that produced increasing returns over time. Suddenly seeing that economics was a high complexity science, his excitement displaced his shock when he saw that the properties of increasing returns-lock-in, unpredictability, tiny events that have immense (nonlinear) historical impact-could explain a new economics, one that rushed forward on the edge of unfolding time, generating dynamic patterns and possibilities as it continuously changed, coalesced, and changed again. Soon, Arthur saw that his vision of increasing returns fit high technology like a glove. High technology was congealed knowledge. Initially, products from the knowledge economy had high fixed costs that gave way to low marginal costs for each copy. Not only was each copy cheaper, but also every copy permitted continual movement down the production learning curve of better, smaller, faster, and cheaper. Customers flocked to a standard technology, producing lock-in, and created a winner-take-all-or-most outcome that could be most lucrative if unstable, given the high rate of displacement by new innovations.
D. Adaptive Schemata. Agents in a CAS respond to their environment, including their interactions with other agents, analyze this information, and formulate a set of conditional (if-then) rules that can be called an adaptive schema. In turn, these agents, using their own adaptive schema based on experiential learning compete and cooperate in a CAS like the stock market. The market winnows out the fit schemata from the unfit as investors profit or go broke. In series of computer simulations, Holland developed a genetic algorithm and proved its fundamental theorem: in the presences of reproduction, crossover, and mutation, almost any compact cluster of genes (a genetic schema) that provides above-average fitness will grow in the population exponentially. One practical outcome of using the genetic algorithm and rules-as-classifiers was its ability to control a simulated gas pipeline, which was demonstrated by Goldberg in a Ph.D. Dissertation that won him a Presidential Young Investigator Award in 1985. The objective in any pipeline system is to meet demand at the end of the system as economically as possible. Pipelines may have thousands of compressors pumping gas through networks of thousands of miles of pipe. Demand for gas changes hourly and seasonally; compressors and pipelines spring leaks, which lower pressure; safety constraints demand that pressure be kept within acceptable limits; and everything changes everything else. Mathematical approaches had failed to optimize such systems, which relied instead on the experienced intuition of engineers. Not only did the rule-based genetic algorithm model achieve expert levels of performance in only 1000 days (runs) of simulated experience, the system was incredibly simple. Its rules formed a default hierarchy, which used a weak general rule, like "Always send a no-leak message," until exceptional conditions required that a stronger, specific rule become dominant, like "if the input pressure is low, the output pressure is low, and the rate of change of pressure is very negative, then send the 'leak' message'. Such an emergent model of complex systems is immensely practical. The evolutionary approach, Holland wrote, "eliminates one of the greatest problems in software design: specifying in advance all features of a problem." Kelley (1994, p. 291) concluded, "Anywhere you have many conflicting, interlinked variables and a broadly defined goal where the solutions may be myriad, evolution is the answer."
E. Self-Organized Criticality. The physicist Per Bak introduced the phrase, "self-organized criticality," as descriptive of the state of a sand pile after piling sand on a tabletop. Whether you drop the sand all at once or a few grains at a time, the resultant sand pile will approximate the shape of a cone on the tabletop. What is important for our purposes is that the resulting sand pile is "self-organized," in the sense that it reached a steady state without anyone explicitly designing or shaping it. It is in a state of "criticality," in the sense that the sand grains on its surface are just barely stable. According to Waldrop (1992, p. 304), "In fact, the critical sand pile is very much like a critical mass of plutonium, in which the chain reaction is just barely on the verge of running away into a nuclear explosion, but doesn't." If someone drops a few more grains of sand on the pile, the outcome is unpredictable: sometimes nothing happens, sometimes cascades of various smaller sizes occur, sometimes a catastrophic avalanche caves in a whole side. Big avalanches are rare; whereas, small cascades are frequent. In fact, the average frequency of a given size of cascade is inversely proportional to some power of its size. This mathematical relationship between frequency and an observed behavior is called a power law. Power laws are ubiquitous in nature, describing the flow of water and electricity, earthquakes and tidal waves, and fluctuations in a stock's price and in indices of stock markets.
Applications of CAS to the Market.
Mauboussin applies these ideas to capital markets. The classical capital market theory provides the context in which security analysis is conducted. Yet, there is an ongoing conflict between financial theorists and practical investors over whether any investor can outperform the market over time. The classical model stresses stock market efficiency, citing rational agents who use all available information, random walks in stock prices, and the failures of chartists and most money managers. The practitioner's model sites the unusual performance of Buffet and his like, the irrationality of investors in emotional responding and group think, sudden shifts in prices, and descriptions of the emotions of the market, as variously excited, jubilant, jittery, and downtrodden.
Believing there is something to be said for both sides, in "Shift Happens," Mauboussin examined the evidence and concluded that a paradigm shift (in Thomas Kuhn's sense) was currently underway, a shift from the classical economic theory of market efficiency and equilibrium toward a new paradigm of the market as a CAS that extends comprehension without undermining all that has come before.
When capital markets are viewed as an equilibrium system responding to linear causes with linear effects, it posits a balance between supply and demand, risk and reward, price and quantity. Stock market efficiency claims that current stock prices reflect all relevant information, that they are not systematically mispriced, which means that the investor will be compensated for his risk and no more. Assuming that current prices reflect all information that is collectively known by investors, random walk posits that changes in prices would come only from unexpected information, which is defined as random. One implication is that the probability distribution of returns will be normal. A second implication is that rational agents are assumed to seek the highest return for a level of risk (means/variance efficiency). Classical capital markets theory assumes that price returns are normally distributed and that agents are rational individually or collectively. It predicts modest trading activity and limited fluctuations in price.
In fact, stock market returns are not normal. They exhibit kurtosis: tails are fatter and the means are higher than expected. The market's booms and crashes generate the higher-than-predicted changes from normality. Non-normal distributions undermine random walk and reduce confidence in statistical techniques that assume normality. Moreover, some series of returns are both persistent and trend-reinforced. Also, trading volume is higher and price changes are greater than predicted. Furthermore, risk and reward are not linearly related via variance. Finally, investors are not ideally rational; they use simplified rules of thumb in inductive reasoning that neither encompasses all information about the market nor other investors' expectations. Not only that, if enough agents adopt decision rules based on the momentum of price activity, they can create a price trend that is self-reinforcing.
Mauboussin argued that CAS accounts for some of the inconsistencies between classical theory and reality. Using the concepts of punctuated equilibrium and self-organized criticality, a period of stability punctuated by rapid change at critical levels, the kurtosis of booms and crashes and high levels of trading activity are predictable and consistent. When trend persistence is found throughout nature, it is not surprising in capital markets. Mauboussin regards the shift from viewing investors as rational agents using deduction to inductive decision makers with bounded rationality as crucial. Inductive decisions will yield prices close to "intrinsic value" if errors are independent. Here, an example will help.
Both Mauboussin and Paul Johnson collect data from their classes at Columbia to illustrate the importance of independent errors in decision-making. Students are asked to make predictive judgments about the winners of Oscars in twelve Academy Award categories. Each puts a dollar into the pot for the winner. The groups modal choices are correct on about eleven of twelve, the average student gets about five correct, and the best individual about nine. A diverse group of agents making independent errors creates a better answer; this best answer in an emergent from the CAS; it illustrates Adam Smith's "invisible hand."
Take-Away for Investors.
1. If there is not a linear link between risk and reward, then owning a stock exhibiting persistence may be less risky that previously believed.
2. If large-scale changes can come from small-scale inputs, then cause and effect thinking is simplistic and counterproductive.
3. Traditional discounted cash flow analysis remains valuable for sorting out investment issues.
4. In a new paradigm world, strategy must incorporate rules about technology lock-in and increasing returns from network effects.
I hope this helps.
Don |
