5 Key Principles of Complexity Economics
Complexity Science, among the most important bodies of theory to have emerged globally in the second half of the 20th Century, can be defined as the study of naturally occurring systems. This makes the discipline difficult to categorize, since systems occur everywhere, and insights from complexity studies have correspondingly been integrated in all of the natural and social sciences. Few fields, however, have been affected quite as thoroughly and profoundly as that of economics.
The relatively young discipline of Complexity Economics does not represent a ‘sub-field’, like Behavioural Economics or Financial Economics. It is more appropriate to see it as a ‘super-field’ (super here meaning ‘above’), as it reframes the entire field of economic studies and offers tools for abstract thinking that were hitherto unavailable to the discipline. Although still largely obscure to the general public, it is enormously consequential, not to mention fascinating, and anyone with even the most casual interest in economics would surely find it rewarding to investigate.
In this article, I will attempt an introduction not to Complexity Science at large (the field is vast and enormously interdisciplinary), but to Complexity Economics specifically.
I will introduce five key principles in Complexity Economics, explaining them as simply as I am able, and through them I will attempt to sketch a conceptual picture of what an economy is and how it works. I will occasionally contrast these principles with the ideas we inherited from ‘traditional economics’ — an imprecise expression that nonetheless spares me getting lost in rabbit holes regarding which economic school held which stance exactly.
I should add as a disclaimer that I am not a scientist or an economist myself, only an aficionado of the topic. I say this not only to temper the claims in this article (if you find an academic paper that contradicts something I say, trust the academic paper), but also to highlight that the topics of Complexity Science are not as ‘complex’ as the name suggests. Instead, they are within the grasp of any educated reader who is willing to look past the occasional jargon, and I hope to prove this right away as I jump into discussing our first principle (mind the jargon!).
Principle #1: Economic agents operate by heterogeneous inductive reasoning
An economy is composed not only of material goods, but of people, companies, institutions and other economic agents interacting with each other as they pursue their respective interests. In traditional economic theory, these agents are generally assumed to be rational and homogeneous.
Those terms call for a little elucidation. Saying that agents are homogeneous doesn’t mean that they are all clones of each other with the same interests, nor even that they all have access to exactly the same information. Rather, agents are homogeneous in the sense that when presented with the same information, they will all interpret it in the same way — the ‘rational’ way.
Now rationality, in this context, doesn’t simply mean ‘being able to reason’. What it means is that agents perform analyses and reach conclusions by a manner of reasoning called deductive logic. This is the logic used by computers, as well as in disciplines like pure mathematics. It reaches conclusions by stringing together the inevitable implications of different statements: if A implies B, and B implies C, then A implies C.
One of the problems with using rational economic agents to model the economy is that real agents (that is to say people, because even companies and institutions almost always trace their decision-making and behaviour to people) think by means of deductive logic only very rarely.
Instead, human beings primarily think by means of inductive reasoning: we see patterns in the world, we form a hypothesis (or various hypotheses) based on those patterns, and we test our hypothesis against events as these unfold. If the hypothesis holds true in our experience, we retain it, and with sufficient corroboration it becomes a belief. Otherwise we discard it, and go on to form a new hypothesis based on whatever new patterns present themselves to us.
This way of thinking is called inductive, as opposed to deductive, because we are not extracting a necessary conclusion from the world before us. Instead, we are projecting a conclusion (one that is heavily influenced by our previous, personal experience) on the world before us. For example I may travel to a different country, see a police agent in a red uniform, and conclude — based on my experience that police agents in my country all wear uniforms of the same colour — that police agents in this country all wear red. As I keep travelling across this country, I will see other police agents and I will be able to verify my hypothesis, which will allow me to either consolidate it into a belief or discard it.
The problems with assuming homogeneous rational agents in theories of the economy were known long before Complexity Science, and indeed many economic theories have adjusted their models accordingly.
In Complexity Economics, the assumption is wholly discarded. Models in this discipline assume agents that a.) operate by inductive reasoning, projecting their conclusions on the world based on their previous experience and constantly updating their beliefs as new experience becomes available, and following from this, b.) are heterogeneous, not only in that their interests and goals may differ, but also in the sense that each interprets the same data differently, because they all have different pre-existing beliefs and experiences.
Principle #2: Market behaviour is unpredictable, but its unpredictability has parameters
We said that the economy is the product of the interaction of its agents. In traditional economic theory, each rational agent finds the arrangement that maximizes their respective interests, until nobody has any further incentive to change what they are doing: this is called the economy reaching equilibrium. If an external factor changes the conditions of the economy, the agents will rearrange themselves in time into a new equilibrium.
In Complexity Economics, however, we have agents that think and behave differently than those in traditional economics, and as a result they create a different kind of regime than equilibrium. What sort of regime might this be?
There is a well-known scenario in Complexity Economics called ‘the El Farol Problem’, based on the name of a bar in Santa Fe (USA), that helpfully illustrates this question. Imagine that every evening, 100 people need to decide whether to go to a bar or not. The bar is pleasant to be in, as long as there is room for everyone. However, if more than 60 patrons are in the bar, the place will be too crowded for anyone to have a good time, and it would be best to go somewhere else. Thus, each person will go if they believe that less than 60 other people are going, and will stay home if they believe that more than 60 others will go.
What makes this problem really tricky is that it’s impossible for more than 60 of the 100 people to ever be right at the same time, no matter how smart they are or what information they are given. This is because, if 61 people believe the bar will not be crowded, they will all go — and the bar will end up being crowded. On the other hand, if 61 people believe the bar will indeed be crowded, they will stay home — and the bar will end up not being crowded.
Even if we endowed our agents with perfect rationality, the El Farol problem has no logical solution, because as soon as enough people find the ‘correct’ answer for any given night, that answer flips and becomes incorrect! The people deciding whether to go out must fall back on induction, based on historical data (‘I remember the bar was always crowded on Friday nights, so I will stay home on Friday nights’).
If iterated over time, would this problem eventually settle on an equilibrium, for example with exactly 60 or exactly 50 people going to the bar every night? Or would the agents create a different regime over time, and what would that regime look like?
The problem was modelled by economist W. Brian Arthur in 1994 in a famous paper. We can see what bar attendance would look like over the first 100 weeks in the following graph.
This graph is remarkable for being neither total randomness nor equilibrium. The mean average of the number of patrons converges always to 60 — that’s what the horizontal line in the graph represents. But the exact number of patrons coming each night is unpredictable, and bounces up and down sometimes dramatically, sometimes very moderately. As importantly, Arthur found that the individual agents themselves never settle on any belief, but constantly change membership from the group predicting a crowded bar to the group predicting a non-crowded bar, and the other way around.
It may not have escaped the more attentive reader that there is a major component of the modern global economy that presents the same exact decision-making problem as the El Farol scenario: it is the stock market, in which traders will make a profit when they make a correct prediction and most other traders make the wrong one. Stock traders must therefore attempt to predict the predictions of other traders, but notice how the El Farol trap springs again: if too many people buy the same ‘good’ stock, this will push its price above its value, turning it into a bad buy, and the other way round if too many people sell. In other words, if too many people hit on the ‘correct’ prediction, this flips to become the incorrect one!
The rational agents of traditional economics would, in due time, make the price of each stock settle on its exact value. But we have already established this is not how real agents behave. What happens instead is that each trader will constantly test new strategies to beat the market; occasionally, one such strategy will prove successful, which will make some other strategies obsolete and force the remaining ones to readapt and evolve. As these strategies change, the market itself will change, opening opportunities for new strategies to become successful, which of course begins the process of transformation again.
While the El Farol problem shouldn’t be seen as perfectly exemplary, it does help to illustrate some qualities of free market behaviour in general. It is not completely random and unpredictable; in the above example, the system’s behaviour converges to an average of 60. It is also not completely predictable — the exact number of patrons who will come on any given night (or of traders buying and selling a given stock) cannot be inferred from any pattern within the data. It has structure, and this can be known, studied and revealed; but it has an in-built degree of uncertainty too.
An important difference between the El Farol bar and the stock market is that the latter is, of course, several orders of magnitude more complex. If you think of the straight line in the El Farol graph as an attractor (that is to say, a value around which the graph’s numbers converge), then a system like the stock market is capable of having multiple attractors existing concurrently, as well as much more complex attractors which don’t necessarily proceed in a straight line but may exhibit all sorts of peculiar trajectories. This, for example, is what the famous Lorenz attractor (the mathematical system from which there originated the popular ‘Butterfly Effect’ concept) looks like:
In the example above, the attractor is composed of two very distinct ‘wings’, and when the system shifts from one wing to the other (changing the behaviour and orientation of the entire system with it), this is called a phase transition — a concept to which we will return later in this article. If you think of the two wings in this attractor as metaphorically representing the bull and bear phases of the stock market, then we would be able to describe the system as being constrained to bull and bear behaviour, but we would never be able to predict at which point it will shift from one to the other. (I stress the word metaphorically in the previous sentence — I am using these terms to illustrate broader concepts, but the El Farol average isn’t rigorously a mathematical attractor, and the Lorenz system does not model the stock market).
Needless to say, the history of the stock market demonstrates that it never reaches a fabled state of equilibrium, but instead periodically undergoes numerous phase transitions, and changes constantly over time as attempts to predict its behaviour end up transforming that same behaviour.
Principle #3: Technology is the skeleton of the economic organism
Traditional economic theory treats technology as something connected to but fundamentally other than the world of production and consumption. Technological innovation may sometimes upend the equations of the economist, but those equations naturally do not and should not be expected to account for technological innovation. Where economists bother to study the effects of technology on the economy, this is done only on a retrospective, case-by-case basis rather than in terms of general principles — in other words, it is only studied in the field of economic history.
Complexity Economics distances itself from this perspective by treating technology not only as an organic part of the economy itself, but as its fundamental, underlying structure. Technology is the skeleton of the economy, so to speak, while production and consumption are the muscle and skin around that skeleton.
This argument lends itself to polemics. The various definitions of ‘technology’ used by Complexity Economists can seem a little arbitrary, and philosophers would no doubt find many ways to question them and break them down. But for the purposes of this article, it’s not really necessary to stray very far from an intuitive, common-sense understanding of what technology is.
What matters rather are the properties of technology, and specifically five of them. Firstly, technology comes into being in response to different needs (generally social needs). Secondly, technology creates new needs (these can be in the form of opportunities or problems alike), calling forth yet more new technologies. Thirdly, technology is created by a combination of existing, lower-level technologies, and every new technology thus formed becomes a component in the creation of some higher-level technology. Fourthly, technology consists of families, that is to say, groups of related technologies (e.g. steam, electrical, digital), and these families then cross over with other families (aeronautical technologies combine with military technologies), creating a complex fractal network. Lastly, new technology implies the obsolescence of old technology — that which it replaces, along with all of its family and dependent families.
These properties are important because they allow us to sketch a working schema for how structural change happens in the economy — the sort of change that takes decades, and which traditional economic theory is ill-equipped to think of. It begins and ends with technological development, and it follows four steps precisely. Let’s go through them.
1. A new technology first enters the economy in response to some need, and gradually becomes available to replace existing technologies that addressed the same need.
2. As the new technology spreads, the old technologies it replaces begin to go extinct; this process spreads to many of their related technologies, creating a chain effect of extinction.
3. The new technology becomes available as a component in the creation of new technologies, creating a chain effect of innovation.
4. The innovation produced by the new technology gives rise to new problems and opportunities — in other words, to new needs.
Naturally, the process is cyclical — needs call forth technologies, which call forth new needs, and so on. And yet the cycle never ‘starts again’ in the sense of returning to its starting point; instead, as each new technology joins other technologies as a component of a future technological family still in the process of formation, a new tier of complexity is added each time.
As well, we usually picture a cycle as something that goes round in a circle. But the geometry of this particular cycle is distinctly fractal: each new technology branches out into a whole series of new connections and discoveries, and each technology that goes extinct brings down with it its entire family (with all of its concomitant economic activity). The result is a constant cascade of change.
Structural change in the economy should be understood as the prices of goods and services adapting themselves to this process of technological evolution. The economy adapts itself to technology — they do not adapt to each other, nor does the process go the other way round (or at least not for the purposes of this particular conceptual model; there are of course many senses in which we can say that ‘technology adapts itself to the economy’ — I did say the subject lends itself to polemics! — but I will leave that discussion for another day).
The implications of this way of thinking about structural change are two, and I’m afraid neither is reassuring.
The first is that the economy is ever-changing. This may seem like a truism, but it matters because in the previous section we looked at how it is possible to model complex systems such as the stock market, and identify qualities and constants that define their structure. This allows for studies of the economy that are hugely illuminating… but also perishable. The complex economic structure that can be identified by computer modelling is bound to change, and so our model will no longer be valid.
The second is, yet again, an element of inherent unpredictability. While it is possible to model the technological steps I described above and account for their unpredictable behaviour mathematically, a simpler and more intuitive demonstration is readily available. It is offered by the concept of technological lock-in, which tells us that if a technology appears early enough, it will prevail even over more efficient alternatives (effectively, a failure of the free market to settle on its optimal parameters). The most commonly-cited example is that of the QWERTY keyboard, which is notoriously a suboptimal layout, and yet which has enjoyed a stranglehold on the market for the simple reason that it came in time for everyone to get used to it.
Since technological lock-in depends on accidents of history far too small to model or predict, and since technological development — as we have seen — is fractal and causes cascades of change, the ultimate direction in which technology — and therefore the economy — will develop is inherently unpredictable. This includes the possibility of catastrophic systemic collapse (a return to a ‘Dark Age’ of some kind), which is not uncommon in complex systems, and which may well be caused by endogenous rather than exogenous causes — changes that happen inherently within the economic system, and not because some meteorite hit it from the outside.
Principle #4: All systems will be gamed
The economy is a system composed of systems. Among the smaller systems that make up the economy are those that are sometimes called policy systems, which are more discrete institutions, or initiatives, or sets of regulations playing out over time. For example, the ‘war on drugs’ can be described as a policy system.
Policy systems usually have an aim, but can also be exploited for other aims. Let’s stick with the above example of the war on drugs. If a cynical police commissioner finds ways of dressing up some of their investigations as ‘drug operations’ when they are actually about something else, and in so doing succeeds in siphoning greater public funds to their department, this is called ‘gaming’ or ‘exploiting’ the policy system. The war on drugs was meant to stop the use of drugs in society, and instead (or at the same time) ends up serving the personal aims of this commissioner.
Exploitation of a system is traditionally interpreted as evidence of a system’s (at least partial) failure, of its structure having cracks. The insight of Complexity Economics is that exploitation is not a bug — it’s a feature of all policy systems. It’s the result of agents constantly testing out and updating their strategies (something in turn made necessary by the constant evolution of the market).
Systems are gamed precisely when agents attempt new strategies and find one that was not predicted by the original policy system, and since — as we have seen above — attempts to predict the behaviour of agents will end up changing that behaviour, it becomes impossible to build a system that will not, in time, be gamed.
If this seems to throw human beings in a negative light — as creatures naturally, inevitably tending towards corruption and greed — it’s worth pointing out that in simulation models, exploitative behaviour emerges with no need to add ‘greed’ or ‘malevolence’ in any of its virtual agents. Systems are gamed not because people are evil, but simply because by their nature these systems create opportunities which evolving agents will eventually stumble upon. It’s the same process that naturally occurs in the open market, except here it seems perverse because policy systems usually have a stated purpose or aim which exploitative behaviour is not aligned with.
I am keeping the discussion of this particular principle relatively short, because it must be said that even in Complexity Science it remains somewhat speculative. Exploitation requires the concept of ‘novel behaviour not predicted by a system’, and this concept is formidably difficult to simulate (which parameters can account for that which is outside of parameters?). But the idea is interesting, consistent with the other insights provided by Complexity Economics, and so it deserves this brief overview at least.
Principle #5: The economy is not a hierarchy but an ecosystem
So far in this article we have sketched some of the typical traits of complex systems as they become manifest in the economy. We have seen how the interaction of agents in the market produces a regime which is both structured and unpredictable, and also how technology develops in fractal patterns, with each technology branching off into certain new branches while pruning others, then those new branches becoming components in branches of a higher level.
Bringing together these principles allows us to draw a sketch of the economy as a whole, which of course is composed of more than just the stock market and technology. Speaking broadly, we can say that the economy is composed of diverse entities, where an entity can be a person, a particular group of people, a governmental institution, a market demographic, a company or group of related companies, a technology or family of technologies, an infrastructural hub or network, a policy system, and potentially many other things as well.
So far, so good. But what can we say about these entities and the system they collectively create? One thing that is necessary to abandon is the old concept of hierarchy — the idea that the economy is organized in such a way, that a few entities at the top consciously determine the behaviour of more entities beneath them, which in turn determine the behaviour of yet more entities beneath them, and so on to the lowest level. This organizational model was at the heart of Marxist economic theory, and represents one of its most important intellectual legacies; indeed, many modern sociological theories about class, race and gender inherited the idea that society is a fundamentally hierarchical organization.
Instead of seeing an entity at one level and others beneath it, it is more appropriate to say that an entity of a certain category is composed of other entities of a different category (exactly like technologies are composed of other technologies coming together). Following our hierarchical instincts, we may be tempted to say that there are entities at ‘higher levels’ of complexity composed of entities at ‘lower levels’ (i.e. simpler). But this is only a half-truth, because while it is the case that any complex entity must be composed of various simpler elements, this work of composition can just as easily happen by taking elements laterally (i.e. from other entities belonging to the same level of complexity) or even from the top-down (a policy system is an entity generated with components from more complex entities, such as the government and the voter base it interacted with).
As well, even when an entity is composed wholly of lower-level entities, this does not in any way imply that it rules them, or otherwise determines their behaviour in any sense other than the purely mechanistic, much like a lion does not tell the bacteria living within it what to do. The traditional traits of hierarchy do not hold.
Because of its infinitely interconnected nature, modelling an entire national economy (much less the world economy) is an impossible task, even in theory. However, much like the principles at work in the El Farol problem are transferable to the greater system that is the stock market, likewise we can deduce some aspects of the massive system that is the economy by looking at the behaviour of smaller complex systems.
So, what are some of the things we know about the economy?
For a start, the economy possesses large internal structures which can be identified and mathematically defined (although only temporarily, because these structures change in time as technological change happens). Think of the ‘attractor’ in the El Farol problem, leading the behaviour of the system to converge to 60, and imagine many such attractors existing within the economy, like pillars around which behaviour organizes itself.
The economy also possesses smaller structures, but these are characteristically impermanent. Not only do they appear and disappear, the times and manner in which they do so are intrinsically unpredictable. Not surprisingly, these smaller structures are much harder to identify and study.
As well as entities, the economy will be composed of strategies, adopted by those entities themselves down to the last individual agent. Taken together, these strategies form one of the many systems-within-the-system of the economy, and they typically arrange themselves in an ecosystem of mutually dependent coexistence. New strategies evolve, others go extinct, in a process of constant change.
The changing strategies within the economy will affect the parameters of the above-mentioned mathematical structures, both large and small. When these parameters have been changed past a certain point of criticality, the system as a whole will undergo a phase transition. This means that the behaviour of the system will change significantly, sometimes even radically, and relatively suddenly.
The nature of a phase transition can take many forms, but there are at least three recurring types of phase transitions that can be defined as inherent to the economy — as in-built to the system and impossible to get rid of definitively. It’s worth pointing out that none of these phenomena are theoretically possible in an economy that converges towards equilibrium.
The first type of phase transition comes in the form of bubbles and crashes, which we can describe together as they are essentially the same phenomenon going in two different directions. This is what happens when the perceived value of assets in the economy dramatically goes up or down in a very short period of time, without necessarily reflecting the true value of said assets. Sometimes bubbles and crashes can be very contained and affect only one particular domain of the economy, while other times — think of the stock market crashes in 1929 and 2008 — they can devastate entire countries.
The second type is the alternation from a long stretch of activity to a long stretch of inactivity, and the other way round. This has some parallels with bubbles and crashes, but rather than perceived value going up and down, it is the volume of trade and/or the productivity of the entities in an economy that changes dramatically.
The third is the rapid propagation of structural change, which is typical particularly of systems composed of densely connected networks, as is our globalized economy. In effect what is rapidly propagating is not structural change itself, but the evolution (the rapid learning from each other) of strategies internal to the economy, which push the parameters of the economy’s structures past criticality, and that leads to structural change. While structural change is always happening in a complex system, occasionally the network will make that change happen so quickly that it deserves to be called a phase transition in and of itself.
All three of these phenomena can be traced to activity within the system that occurs neither at the micro- nor at the macro-level, but at the meso-level — the one in between, where all of the most interesting manifestations of time and chance happen, and where Complexity Economics finds its true bread and butter.
Embracing the inevitability of these phase transitions has conceptual consequences that are far from trivial. When such transitions occur in the real world, we so often respond to them by calling for some form of systemic change. Following the 2008 financial crash, for example, it was not uncommon to hear people arguing that the crash demonstrated how ‘the capitalist system’ was fundamentally broken, that it led to people losing their livelihoods and social disasters of all sorts. But the study of complex systems tells us that such crashes are as inevitable as they are unpredictable, no matter how wildly we change the parameters of the overall system.
This is not to say, of course, that when disastrous phase transitions occur, nobody should be held accountable. The individuals and banks whose speculation games led to the crash in 2008 should have faced full legal consequences and been compelled to offer reparations (which they weren’t), because it was their actions which led to that particular crash, even if no system can prevent periodic crashes from happening.
Equally, nothing I have said in this article suggests that economic reform is not necessary or desirable. Quite the contrary. It is precisely because phase transitions are inevitable, that building safeguards against their consequences is very much worthwhile — or to put this differently, if we can’t build an unsinkable ship, then that’s exactly why we should be putting lifeboats on it.
That’s what truly effective ‘systemic change’ looks like: not trying to make a new ship that cannot sink, but adding lifeboats to the ship we’re already on.
Conclusion: The Future of Economics
The old debate on whether economics qualifies as a science has so far been determined less by the nature of the economic discipline itself, than by the fact that it lagged markedly behind all the other sciences in its theoretical development. Not only has economics struggled to model things like structural change (relegating its study to the field of economic history instead), but the uncertainty of economic matters was dismissed as an embarrassment, as something that disproved its scientific nature.
And yet the history of science in the 20th Century has been characterised by an embrace of uncertainty. Mathematics first, and then physics, chemistry and biology, all came to accept that there can be no knowledge without a measure of inherent uncertainty, and that some processes will by their very nature always be unpredictable or unmeasurable.
Complexity Economics has been hard at work to bring that maturity to the discipline of economics, demonstrating that the unpredictability of the systems it studies is exactly what makes them most like the objects of scientific inquiry. We may never be able to predict the caprice of human beings or the Butterfly Effect of technological development, but our theories are not any less scientific for that.
While the paradigms I have discussed in this article have been making waves in academia since the 1980s, they are still being assimilated by culture at large. In this process, we have seen some memorable moments. The vocabulary of Complexity Science started percolating into broader culture at least as far back as the 1990s, with the release of the book and the film of Jurassic Park and its colourful take on Chaos Theory (the conceptual precursor to Complexity). Terms that once existed only as academic jargon are now commonplace even in media created for children (the word ‘fractals’, for instance, appears in the famous Let It Go song from Disney’s 2013 film Frozen).
More recently, the publication in 2007 of Nassim Taleb’s global bestseller The Black Swan marked a point in which not just the vocabulary but many of the key concepts of Complexity Economics became available to the broader public. Popular literature on economics has since found it increasingly easy to integrate the learnings of this once abstract and remote field.
Cultural change is no faster than structural change, and organizing principles we have long taken for granted — like rational agents, economic equilibrium, or the hierarchical nature of the economy — may take a long time yet to die out. But they are due for extinction, and the seeds of their replacements have already been sown. Of the many small flowers announcing the coming spring, this article hopes to be one.