Doyne Farmer - Macroeconomics From the Bottom Up

About the Interview

In 2006, the Fed asked its macroeconometric model what would happen if house prices dropped by 20%. The model projected the past into the future and said: "Not much." Well, the financial crisis proved it wrong. Meanwhile, DSGE models, the main alternative up to this date, do not feature financial institutions; "They are not even good enough to be wrong," says Doyne Farmer. That's why Farmer and his team are developing an agent-based model, of the housing market first and of the entire economy next, to mimic the current financial crisis. The team collects data on actual people to calibrate a rich model with millions of interacting agents. This is a bottom-up approach to macroeconomics -- this is new economic thinking.

About Doyne Farmer

J. Doyne Farmer is a professor at the Santa Fe. His main interests are complex systems, with applications to financial markets and techological innovation. At Los Alamos National Laboratory he was an Oppenheimer Fellow and founded the Complex Systems Group in the Theoretical Division. He then founded the Prediction Company, which does fully automated quantitative trading for UBS. Full profile



Dear Dr. Farmer, We at Perm Stae University (Risk Lab, University & PROGNOZ initiative) have a similar project to study stock market of Russia and India (emerging) by using agent based models. We have the data and are in process of implementing the models. We are doing it through recverse way, which is extaly the opposite the way you are planning to do? I just wanted to know, Are we going in right direction by doing so? regards pankaj


I'm surprised no-one has had a go at this already. Not surprised that this has come from a physics and dynamic systems person.

I've been looking a little at economic systems interrelationhips, utilising causal diagrams, just to get a feel for the causes and effects, and feedback that can occur in an economy and the inherent complexity(notionally of any size), initially in the context of business improvement behaviour. DF is right on the assumptions one has to make when you get to modelling individual 'transactions', whether linear or non-linear, but I'm sure this level of modelling will yield more insight than the traditional macro-level models. Initially, it gets us to ask the right sort of questions.

The nature of complex systems is inherently chaotic e.g. a perturbation in the behaviour of one small agent (input) can, under particular conditions, trigger a large significant outcome. Of course we see this in real cases.
I look forward to seeing some interesting work.
Good Luck.


I see parallels to the information/knowledge asymmetries paradigm. Given the knowledge asymmetries (as postulated by 2001 Nobelists in Economics Stiglitz, Spence, Akerlof), a "real knowledge divorce" between the two levels in economics and the resultant inadequacies at micro- and macro-levels in the political-economic systems in the global economy cause a new type of crises which cannot be easily contained, let alone prevented, on the strengths of the EMH based self-correction mechanisms. This problem is brought about mainly by the speed of capital and information flows in the global (digital) economy. It follows therefore that on top of understanding the real behavior of real economic agents (and their dynamic interactions in complex systems) what we need is new institutions that help speed up the necessary "digestion" of knowledge (to prevent "rubbish in, rubbish out" modeling syndrome) so that policy responses are much better informed and adequate than before. Such knowledge management institutions (e.g. college of macroprudential regulators, etc) would make very good use of the new understanding of the economic principles (e.g. real agent based) and in the longer run provide a much less expensive solution than correction via the great depression. Very interesting and new thinking indeed!


RE: I'm surprised no-one has had a go at this already.

Not exactly a big surprise. Mr. George Soros has been proposing parts of this approach in his "Reflexivity Theory". Long before the onset of this Great Recession I wrote (in the context of advise to emerging markets) about the hitherto unfamiliar micro-macro troubles in economics, created by the new global (digital) economy and the new nature and speeds of international capital flows; nobody was interested, however. Prof. McKinnon of Stanford wrote about similar problems as early as 1971! Btw, the Tobin tax proposal, developed a long time ago, also alludes to these aspects of analysis. I can multiply the examples but you get the point.


Right at the beginning (almost two years ago) I have submitted exactly such a project to INET but this was rejected... «Too many submissions».
I am a Sociologist by training with a life long experience in computers and programming. I have in fact continued my project, now with the help of a colleague from the doctorate program in complexity sciences at ISCTE - The Lisbon University Institute.
It all started with the development of a computer algorithm that I developed some 30 years ago to demonstrate the «marxian» «law of the falling tendency of the rate of profit» following on the ideas of Prof. Ronald Meek.
It was presented at a congress and it showed a clear tendency for the systemic growth of unemployment. And the tendential fall of the rate of profit. Just like the «proof» advanced by Okishima, the rate did go up during the first cycles (years), but then flattens out and starts to fall...


Love what Doyne and team are doing! This is probably some of the most important work being done in the field today, and I'm very encouraged to see that INET is supporting the initiative. In general, I'm a huge fan of SFI's approach, especially in the vein of John Holland and Brian Arthur. As an educator and a researcher, in my opinion this is the kind of work that will very much interest and inspire today's young minds.

Doyne, is there room for others to assist the project? I'd love to see some firm-level or project-level models as well... I think it would be really interesting and useful to model the RBV approach (resource based view of the firm ie Wernerfelt, Barney etc) or the Knowledge based view of the firm (Grant 1996, Grant & Spender etc). Both of these perspectives have heterogeneous resources at their core. Seems like it would fit nicely with the modelling of heterogeneous agents...


The project led by Doyne Farmer and other projects developed by specialists in complexity studies (complexity science as it is even claimed), are quite fascinating but we must remember about the limitations of that approach which are well-known. There were already quite a few applications of complexity models in economics, finance and social sciences.

Agent-based models (Complex Adaptive Systems – it is not completely equivalent approach), not always help in “direct” predictions, especially if it concerns systems including human actors. They allow to identify “emerging” properties which may not be treated as identical with possible course of events in real systems. They can help prediction in direct and indirect way. They can directly help in making predictions/scenarios based on the outcome of modeling. Sometimes the emerging properties are very remote from any routine ways of reasoning so the agent-based models may help in identifying the courses of events which are “unthinkable”. It makes them a valuable heuristic instrument by enriching our understanding and imagination concerning complex social phenomena.

Predictions with broadly defined complexity models are more adequate for simpler cases and some good results have been achieved by econophysics in finance, e.g. modeling transactions on the Forex. So we should see this approach not as the panacea for improving predictions but as a valuable supplementary instrument of economic thinking.

Since Doyne Farmer is one of the “Founding Fathers” of complexity studies and highly competent in that field, we may expect that his research team will throw more light on the causes of the housing market bubble but it will be only another example of backcasting or retrodiction which indirectly could be very useful in making predictions. Expectations for any extended possibilities of direct prediction with agent-based models in economics/finance/social sciences should be rather cautious.


This is brilliant, if dangerously ambitious. Agent-based modeling of dynamic systems has been around in economics for some time, but it's a neglected stepchild. See Agent-Based Models in Economics and Finance, Luna and Perrone, eds., published in 2002.

Modeling the financial crisis is going to require a real focus on risk and how mortgage risks were transferred through the financial system. Economic behavior is largely determined by perceptions of risk and the mistakes occurred where people were not aware of risk and how the markets were separating risk from reward. An expert on risk theory seems essential for this project. I believe uncertainty and risk is also key to understanding non-linear economic dynamics because it informs us about how agents behave in networks.


What is your point exactly? Complexity Sciences have already shown us they are useful to understand the physical, biological and social reality. As Dr. Farmer pointed to us, standard economics-based models do not work for a number of reasons. Please compare the amount of resources devoted to mainstream economics and complexity. These guys clearly deserve an opportunity, it is only fair!


Just watched the videos on the Complexity Theory interviews. Fantastic stuff.

I was also inspired by the revolutionary spirit and drive Mr Farmer exhibited.

I think agent based modelling is secondary. For example, I think the complex system approach is also being taken by Prof Steve Keen, though he is modelling things differently.

Key is that dynamic non-linear systems with emergent properties are antithesis to mainstream economics and this is a real tragedy. Another key point is that the financial system (along with its media and telco auxiliaries) should be regarded as the control system but is not. Good governance needs the collective awareness that finance is governance and that this is a complex system. Who and what will shape it once that awareness is in place is another question.


I find that the way of reducing the complex situation to an aggregate one for a number of different idealized entities is a simpler and more satisfactory for understanding how the system as a whole actually works. (See my model in Wikimedia, Commons, Macroeconomics, DiagFuncMacroSyst.pdf
I also believe that this is the right direction to go. It is important to see (after the introduction of a matrix representation) that it also allows the model to be greatly explanded so that the agents are represented too. However, in my opinion there is little advantage in doing this since the mechanism behind the overall dynamics is already in place.


The subject of complexity theory is likely to become unintelligable unless some kind of simplification is introduced. Thus by analogy in physics, the molecules of a confined gas can have general properties relating their density, temperature and pressure, even though today we also know a lot about the distribution of their individual energies. Such approximations are the sole means for understanding about how the system works as a whole.

When as in macroeconomics we have much more than one kind of agent (e.g. firms, banks, capitalists, producers, etc), they cannot be lumped together and do not behave in a similar manner. The example given in the talk about housing purchases is one of these separate kinds of activities and to add to the complexity of it for all of these kinds of activities is not going to help us understand the way the "big picture" works.

In my analysis of the macroeconomics system for a closed economy I find that only 6 (idealized and aggregated entities) having no more than 19 different kinds of actions, are sufficient to properly and fully describe the whole system in the most general terms.

So I suggest that complexity theory should not be treated as indivual agents actions, but better as the averages of various kinds of things we commonly share and observe within the whole shebang.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <blockquote> <br> <cite> <code> <dd> <div> <dl> <dt> <em> <h2> <h3> <h4> <img> <li> <ol> <p> <span> <strong> <sub> <sup> <table> <tbody> <td> <tr> <ul>
    Allowed Style properties: display, float, height, margin, margin-bottom, margin-left, margin-right, margin-top, width
  • You may insert videos using embed codes like these:
    • [video_large:KoqLu5CKx-o]
    • [popupvideo_mini:KoqLu5CKx-o right]
    • [lightboxvideo_mini:KoqLu5CKx-o]
    • [text_popupvideo:KoqLu5CKx-o nostart noicon|Click here to open the video.]
    • [text_lightboxvideo:KoqLu5CKx-o|Open this video in a lightbox.]
    To learn more, please click on the "More information..." link below.

More information about formatting options