Complexity Theory and Evolution in Economics

This book looks interesting:

• David S. Wilson and Alan Kirman, editors, Complexity and Evolution: Toward a New Synthesis for Economics, MIT Press, Cambridge Mass., 2016.

You can get some chapters for free here. I’ve only looked carefully at this one:

• Joshua M. Epstein and Julia Chelen, Advancing Agent_Zero.

Agent_Zero is a simple toy model of an agent that’s not the idealized rational actor often studied in economics: rather, it has emotional, deliberative, and social modules which interact with each other to make decisions. Epstein and Chelen simulate collections of such agents and see what they do:

Abstract. Agent_Zero is a mathematical and computational individual that can generate important, but insufficiently understood, social dynamics from the bottom up. First published by Epstein (2013), this new theoretical entity possesses emotional, deliberative, and social modules, each grounded in contemporary neuroscience. Agent_Zero’s observable behavior results from the interaction of these internal modules. When multiple Agent_Zeros interact with one another, a wide range of important, even disturbing, collective dynamics emerge. These dynamics are not straightforwardly generated using the canonical rational actor which has dominated mathematical social science since the 1940s. Following a concise exposition of the Agent_Zero model, this chapter offers a range of fertile research directions, including the use of realistic geographies and population levels, the exploration of new internal modules and new interactions among them, the development of formal axioms for modular agents, empirical testing, the replication of historical episodes, and practical applications. These may all serve to advance the Agent_Zero research program.

It sounds like a fun and productive project as long as one keeps ones wits about one. It’s hard to draw conclusions about human behavior from such simplified agents. One can argue about this, and of course economists will. But regardless of this, one can draw conclusions about which kinds of simplified agents will engage in which kinds of collective behavior under which conditions.

Basically, one can start mapping out a small simple corner of the huge ‘phase space’ of possible societies. And that’s bound to lead to interesting new ideas that one wouldn’t get from either 1) empirical research on human and animal societies or 2) pure theoretical pondering without the help of simulations.

Here’s an article whose title, at least, takes a vastly more sanguine attitude toward benefits of such work:

• Kate Douglas, Orthodox economics is broken: how evolution, ecology, and collective behavior can help us avoid catastrophe, Evonomics, 22 July 2016.

I’ll quote just a bit:

For simplicity’s sake, orthodox economics assumes that Homo economicus, when making a fundamental decision such as whether to buy or sell something, has access to all relevant information. And because our made-up economic cousins are so rational and self-interested, when the price of an asset is too high, say, they wouldn’t buy—so the price falls. This leads to the notion that economies self-organise into an equilibrium state, where supply and demand are equal.

Real humans—be they Wall Street traders or customers in Walmart—don’t always have accurate information to hand, nor do they act rationally. And they certainly don’t act in isolation. We learn from each other, and what we value, buy and invest in is strongly influenced by our beliefs and cultural norms, which themselves change over time and space.

“Many preferences are dynamic, especially as individuals move between groups, and completely new preferences may arise through the mixing of peoples as they create new identities,” says anthropologist Adrian Bell at the University of Utah in Salt Lake City. “Economists need to take cultural evolution more seriously,” he says, because it would help them understand who or what drives shifts in behaviour.

Using a mathematical model of price fluctuations, for example, Bell has shown that prestige bias—our tendency to copy successful or prestigious individuals—influences pricing and investor behaviour in a way that creates or exacerbates market bubbles.

We also adapt our decisions according to the situation, which in turn changes the situations faced by others, and so on. The stability or otherwise of financial markets, for instance, depends to a great extent on traders, whose strategies vary according to what they expect to be most profitable at any one time. “The economy should be considered as a complex adaptive system in which the agents constantly react to, influence and are influenced by the other individuals in the economy,” says Kirman.

This is where biologists might help. Some researchers are used to exploring the nature and functions of complex interactions between networks of individuals as part of their attempts to understand swarms of locusts, termite colonies or entire ecosystems. Their work has provided insights into how information spreads within groups and how that influences consensus decision-making, says Iain Couzin from the Max Planck Institute for Ornithology in Konstanz, Germany—insights that could potentially improve our understanding of financial markets.

Take the popular notion of the “wisdom of the crowd”—the belief that large groups of people can make smart decisions even when poorly informed, because individual errors of judgement based on imperfect information tend to cancel out. In orthodox economics, the wisdom of the crowd helps to determine the prices of assets and ensure that markets function efficiently. “This is often misplaced,” says Couzin, who studies collective behaviour in animals from locusts to fish and baboons.

By creating a computer model based on how these animals make consensus decisions, Couzin and his colleagues showed last year that the wisdom of the crowd works only under certain conditions—and that contrary to popular belief, small groups with access to many sources of information tend to make the best decisions.

That’s because the individual decisions that make up the consensus are based on two types of environmental cue: those to which the entire group are exposed—known as high-correlation cues—and those that only some individuals see, or low-correlation cues. Couzin found that in larger groups, the information known by all members drowns out that which only a few individuals noticed. So if the widely known information is unreliable, larger groups make poor decisions. Smaller groups, on the other hand, still make good decisions because they rely on a greater diversity of information.

So when it comes to organising large businesses or financial institutions, “we need to think about leaders, hierarchies and who has what information”, says Couzin. Decision-making structures based on groups of between eight and 12 individuals, rather than larger boards of directors, might prevent over-reliance on highly correlated information, which can compromise collective intelligence. Operating in a series of smaller groups may help prevent decision-makers from indulging their natural tendency to follow the pack, says Kirman.

Taking into account such effects requires economists to abandon one-size-fits-all mathematical formulae in favour of “agent-based” modelling—computer programs that give virtual economic agents differing characteristics that in turn determine interactions. That’s easier said than done: just like economists, biologists usually model relatively simple agents with simple rules of interaction. How do you model a human?

It’s a nut we’re beginning to crack. One attendee at the forum was Joshua Epstein, director of the Center for Advanced Modelling at Johns Hopkins University in Baltimore, Maryland. He and his colleagues have come up with Agent_Zero, an open-source software template for a more human-like actor influenced by emotion, reason and social pressures. Collections of Agent_Zeros think, feel and deliberate. They have more human-like relationships with other agents and groups, and their interactions lead to social conflict, violence and financial panic. Agent_Zero offers economists a way to explore a range of scenarios and see which best matches what is going on in the real world. This kind of sophistication means they could potentially create scenarios approaching the complexity of real life.

Orthodox economics likes to portray economies as stately ships proceeding forwards on an even keel, occasionally buffeted by unforeseen storms. Kirman prefers a different metaphor, one borrowed from biology: economies are like slime moulds, collections of single-celled organisms that move as a single body, constantly reorganising themselves to slide in directions that are neither understood nor necessarily desired by their component parts.

For Kirman, viewing economies as complex adaptive systems might help us understand how they evolve over time—and perhaps even suggest ways to make them more robust and adaptable. He’s not alone. Drawing analogies between financial and biological networks, the Bank of England’s research chief Andrew Haldane and University of Oxford ecologist Robert May have together argued that we should be less concerned with the robustness of individual banks than the contagious effects of one bank’s problems on others to which it is connected. Approaches like this might help markets to avoid failures that come from within the system itself, Kirman says.

To put this view of macroeconomics into practice, however, might mean making it more like weather forecasting, which has improved its accuracy by feeding enormous amounts of real-time data into computer simulation models that are tested against each other. That’s not going to be easy.


12 Responses to Complexity Theory and Evolution in Economics

  1. Todd Trimble says:

    Insubstantial and irrelevant comment: I like the cover. It reminds me of one of those Oxherding Pictures from the Chan/Zen tradition of Buddhism, specifically the one which is traditionally number 8 in the series (“Bull and Self transcended”) — see

  2. Blake Stacey says:

    That does sound like an interesting book! It’s the sort of thing I was all about, in a previous research life.

  3. Eugene says:

    Thank you for posting the links. The book is interesting and so is Evonomics.

  4. arch1 says:

    The Agent Zero approach is exciting. I hope this model building initiative gets traction and thrives. As a layperson I was puzzled by one thing in the Epstein/Chelen article:

    Does anyone know why Agent Zero’s current skeletal equations allow for an agent’s “disposition to act” to be influenced by other agents’ disposition to act, but not by other agents’ actual actions?

    The article highlights this as crucial, and say that “Despite suspending [this] assumption central to the literature on social transmission … the Agent_Zero model is show to credibly generate a panoply of collective phenomena…”

    But I don’t see any explanation why the modelers chose to assume that others’ actions per se have zero influence (in the real world, by contrast, others’ actions superficially seem both more observable and more influential than mere dispositions).

    • John Baez says:

      arch1 wrote:

      Does anyone know why Agent Zero’s current skeletal equations allow for an agent’s “disposition to act” to be influenced by other agents’ disposition to act, but not by other agents’ actual actions?

      I don’t really know, and I agree this is unrealistic, but I have a theory: it’s easier to write a program where “disposition to act” directly affects “disposition to act”, because they are both variables of the same kind. To get actions to affect dispositions to act, one needs a way to translate actions to dispositions.

      A more realistic model might include this. Of course if we talk to someone we can tell them our dispositions, but humans as well as animals who don’t talk also modify their dispositions based on actions!

  5. Thanks for posting about this! I will add it to my reading list!

    As someone who did a computer science thesis on artificial neural networks, with a blog about personal finance and a interest in understanding human decision making, this is SUPER interesting to me.

    I think it’s interesting that Kate Douglas says, “To put this view of macroeconomics into practice, however, might mean making it more like weather forecasting” because we tend to joke that the weather forecasters are always wrong, or that they have the best/worst jobs because they’re just guessing. People don’t want to think of economists like weathermen because when the weatherman is wrong, worst case you have to walk from you car in the rain. When the economists get it wrong, worst case you lose a lot of money!

    • John Baez says:

      If people are more critical of weather forecasts than economic forecasts, it could be because more people use weather forecasts on a day-to-day basis, and can easily, rapidly tell whether a forecast was correct. Economic forecasts tend to be more long-range and a bit harder to check.

    • Giampiero Campa says:

      I think in some sense weather forecasting is easier than economics forecasting because in the former you:

      1) know the underlying physics (at every level of the abstraction hierarchy).

      2) have really lots of data and measurements to tune your parameters

      i think none of the above condition is satisfied in economics, or at least not at a comparable level.

      Anyways, it would be nice if one day we could have economic models as detailed as suggested. I am not sure we could “understand” them in the same way we can understand and teach simpler models, but, we could definitely use them to run simulations and test hypotheses related to how individual behavior affects, in aggregate, the whole economy. And it would also help understanding and verifying the assumptions on which simpler models are based.

      • @whut says:

        Giampiero Campa says:
        26 May, 2017 at 12:33 am
        “I think in some sense weather forecasting is easier than economics forecasting because in the former you:”

        What’s much easier than weather forecasting is climate forecasting. The patterns in climate are deterministic and driven by forces that are solidly locked in place, mainly based on the lunisolar orbit — not much different that the ocean tidal cycles.

        A reminder that on the Azimuth Project forum we are making progress in describing the apparently “chaotic” ENSO climate behavior as a straightforward solution to a delayed differential equation. Here the well known long-period tidal cycles reinforced by a seasonal stimulus govern every peak and valley observed.

        It’s actually a simple system obscured by the complexity of having to work past a non-linear DiffEq. None of the standard linear models help much in deducing the pattern, which likely explains why the pattern has remained hidden for so long. This is stuff I presented at the AGU last December, but over the last few months, the model has crystalized into a more concise form.

        BTW, that sketched “O” on the cover of the book “Complexity and Evolution” is coincidentally also called Enso — which means circle in Japanese

        “The ensō symbolizes absolute enlightenment, strength, elegance, the universe, and mu (the void).”

        It’s surely a sign for us to work on the stuff that is solvable!

        Paul Pukite

You can use Markdown or HTML in your comments. You can also use LaTeX, like this: $latex E = m c^2 $. The word 'latex' comes right after the first dollar sign, with a space after it.

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.