
The Flux by Epistemix
Welcome to The Flux - where we talk data, decisions, and stories of people asking the what-if questions to create an intentional impact on the future.
The Flux by Epistemix
From Chaos to Clarity with Doyne Farmer: Why Agent-Based Models Matter Now
On this episode of The Flux, agent-based models (ABMs) have long lived on the fringes of mainstream economics but that’s changing. Doyne Farmer breaks down how advances in computing and data are fueling a new era of bottom-up modeling. He explains where ABMs outperform traditional models, why certain fields (like epidemiology and traffic systems) adopted them earlier, and how economics is finally catching up. We explore the structural and behavioral complexity that ABMs are uniquely suited to handle and what it will take to shift institutional thinking in policy and business.
Welcome to The Flux, where we hear stories from people who have asked what if questions to better understand the world and talk about how data can help tell stories that impact decisions and create an intentional impact on the future. This is your host, John Cordier, CEO at Epistemix. In a world where the flux capacitor from back to the future does not yet exist, people have to make difficult decisions without always knowing how the future will play out.
Our guests are people who've taken risks, made decisions when uncertainty was high, and who have assisted decision makers by using data and models. We hope you can turn lessons from our podcast into foresight, so you or your organization can make better decisions and create an intentional impact for others.
Alright. Welcome to The Flux, where we get to hear stories from people who have asked the "what if" questions to better understand the world and learn how data can help tell stories that impact decisions and create intentional impact on the future. This is your host, John Cordier. Today I’m joined by Doyne Farmer.
Doyne, an American complex systems scientist and entrepreneur, has pioneered many fields that have defined the scientific agenda of our times things like chaos theory, complex systems, artificial life, even some early wearable computing. Currently, Doyne is Director of Complexity Economics at the Institute for New Economic Thinking at Oxford, and the Chief Scientist and one of the co-founders at Macrocosm. Doyne, welcome to The Flux. Glad to have you on.
Doyne: Thank you. Very nice to be here.
John: Yeah, absolutely. You came out with a book earlier this year, building on a couple decades' worth of thinking. I’m going to read a section from the inside flap so that listeners have some context for where you’re coming from:
"Using big data and even more powerful computers, we can for the first time apply complex systems science to economic activity, building realistic models of the global economy. The resulting simulations and emergent behavior form the cornerstone of complexity economics. This new science will allow us to test ideas, make significantly better economic predictions, and ultimately create a better world."
So Doyne, maybe give a little background on how you landed on complexity economics and how the term came out of a lot of the agent-based modeling work you've done.
Doyne: How did I personally end up doing complexity economics? Like most things, purely by chance, I would say. I've always been interested in complex systems that part wasn’t by chance. I’ve been interested since high school. But I also always wanted to do something entrepreneurial. I didn’t want to just go around the academic wheel 50 times and then die.
My childhood friend, Norman Packard, and I ran one of the early quantitative trading firms, Prediction Company. During that time, I learned a lot about finance. I started reading the literature and felt like the basic ideas were wrong. After eight years which was plenty of time doing that I had to decide what to do with my life. I was torn between systems biology and applying complex systems to economics and finance. I chose the latter, and everything followed from there.
John: Yeah. So, kindred spirits in the belief that when things aren't looked at from a complex systems or bottom-up view, you're not only going to get bad predictions, but your understanding of the subject will be limited. Our work started in public health, and you took the approach in economics. What were some of the macro or microeconomic principles that just didn’t make sense to you and took you off in a different direction?
Doyne: When I first started back in the '90s, I was reading the finance literature. At the time, the dominant paradigm was efficient markets. If that theory was right, then what we were doing had to be a scam. But we set out to disprove efficient markets theory, and I think we did so successfully as did others around that time.
The theories back then assumed markets were perfectly efficient or that everyone knew the correct valuation of things, which just wasn’t true. The fact that there were wide disagreements is why we trade so much.
That was the first flaw I identified. Then I started thinking more about what was really happening in markets how people actually behave. Ecology turned out to be a better metaphor. Financial markets are ecosystems filled with people using specialized trading strategies. These strategies interact with one another, and each generates returns that may feed or starve other strategies. So, I developed a theory I called Market Ecology. It treated markets as ecosystems of specialized players interacting through the market.
That wasn’t a theory welcomed by the mainstream. They believed markets were efficient, assumed equilibrium, and thought everything was fairly priced. But our trading made a living off deviations from those assumptions.
John: Mm-hmm. For some of our first-time listeners, can you explain these interactions whether between agents or firms and provide more context?
Doyne: Sure. In this context, agents would be investment firms or traders. Their interactions happen because every trade affects the price of an asset. If I buy, the price goes up; if I sell, it goes down. The price you see is a cumulative result of all trades.
So everything every participant does affects everyone else. It’s not like there's a threshold below which you don't impact anything. Everyone has some effect.
From theory we've developed since the '90s, we now understand that market impact is a nonlinear function. Surprisingly, little players can have disproportionately high impact.
When we were trading, one of our close competitors D.E. Shaw unplugged their trading system for nine months. During that time, the slope of our P&L doubled, and it went back to normal when they came back. So we were clearly affecting each other.
Other market activity also acted as a "food source" for both of us. The strategies we were using fed off portfolio balancing behavior. People balancing their portfolios would cause market moves that generated signals we could exploit.
John: So finance was a natural place for you to work, but where else does this type of collective behavior in systems get overlooked? You’ve looked at the economic side; we’ve looked at public health. Where do you see this theory being underutilized but important going forward?
Doyne: Economics and finance are still the biggest examples, because those disciplines have strongly resisted this approach.
But it's been more widely adopted in other areas. For example, in traffic modeling and epidemiology, agent-based models are now standard. Also in battlefield simulations. Even video games are essentially agent-based models. Some of the largest software systems ever built are for video games. It's an interesting reflection on our priorities as a species.
In the social sciences, adoption is more uneven. Political science has some agent-based models, but it's not the focus. Anthropology has used them to good effect. Sociology has used them, but probably not enough. Psychology is a tougher fit since it's more about individuals than systems.
John: So what makes agent-based modeling get adopted in some areas but not in others? Is it the use case, available data, modeling challenges?
Doyne: It’s a mix. Agent-based modeling is more obviously useful when there are a lot of structural interactions and the behavioral components are relatively simple.
Take traffic, for instance. Drivers follow simple rules. You don’t need a sophisticated behavioral model to understand traffic jams or how things change when you add a lane.
Epidemiology sits in the middle. Disease transmission is reasonably straightforward if you understand the disease. Behavioral shifts like people becoming cautious are harder to model, but still doable.
In economics, I think the decision-making part is not as hard as people think. What’s harder is modeling the economy itself: getting all the accounting in place and tracking everything. But that’s what we’re working on.
John: Looking at the structure of the economy individuals, firms, regulatory pressure what are the components you and your team are currently building to understand how small behavior shifts could lead to economic shocks?
Doyne: Our central goal is to make better conditional predictions of what the economy will do under various policy choices. We also want to advise individual firms on strategies how a policy change or competitor move might affect them.
Most firms plan based on assumptions about what their competitors will do, which are often vague. We simulate everyone including competitors so we can better predict that competitive behavior. That’s useful.
We also want to look at externalities like carbon emissions and sustainability. We want to help regulate economies for both prosperity and environmental protection.
John: You mentioned Raj Chetty and the Opportunity Atlas, where a lot of the work has been descriptive rather than structural. What are some of the data or structural elements you're building toward to model the system more holistically?
Doyne: Traditional economic models are limited because they rely on optimization and simplification. Heterogeneous agent models like HANK are an improvement, but they’re still very stylized.
With agent-based models, we can build synthetic populations that incorporate age, education, race, geography whatever’s important and simulate the outcomes for those different people.
That lets us test hypotheses about inequality and social mobility. For example, I hypothesize that the economy is often demand-limited not investment-limited. That is, growth is held back not by a lack of capital, but by a lack of consumer spending. If that’s true, then redistribution isn’t just good for fairness it’s good for growth.
John: That’s a compelling point. Even when everyone gets the same amount say, through UBI you can still see wealth concentrate quickly. Are there others you’d call out who are also championing this type of modeling?
Doyne: There are several. Giovanni Dosi in Pisa. Herbert Dawid in Germany. Rob Axtell at George Mason. Stefan Thurner at the Complexity Science Hub in Vienna. Jean-Philippe Bouchaud in Paris. These are all important voices in this space.
John: What will it take for this approach to become mainstream? What's the inflection point?
Doyne: The inflection point will be when we beat standard economic models on problems that everyone agrees are important. Once that happens and we’re getting close the field will take off.
John: Are there examples from other fields where agent-based modeling reached that kind of inflection?
Doyne: Yes. In epidemiology, it happened after Dick Cheney pushed for bioterror preparedness. That funding helped create sophisticated models, which became the new standard. We haven't had that in economics yet, but we’re working on commercially viable applications that can push the field forward.
John: One of those areas is energy. Can you talk about what you're doing there?
Doyne: We’re building a global energy investment model that simulates every firm and asset 30,000 firms, 150,000 assets. Right now, we’re doing electricity in Canada, oil and gas in the US and Canada, and oil globally. Soon, we’ll model the entire global system.
We’re confident because we have detailed microdata. We can calibrate the model and close the system. We can test what firms could have done better historically and forecast where they're likely to hit trouble.
My fantasy is to advise a company like ExxonMobil and show them they’ll go bankrupt by 2032 if they don’t change their strategy. Whether that’s true or not, we’ll find out. But that’s the kind of scenario we’re testing.
John: You touched on helping companies see themselves in the model. How do you get firms to shift from "this is our data, and that’s all that matters" to recognizing they’re part of a larger system?
Doyne: Some already realize that they just lack the tools. When you can literally see yourself in the model, watch what happens to you, that gets your attention. People are selfish, and that helps them care.
John: Some modelers, like Josh Epstein, lean into "just build the model" versus starting from data. Where do you fall?
Doyne: I think you should look at the data first and build the model to take advantage of it. That gives you a better structure and makes your model more useful. We finally have the data and compute to make this viable.
John: What else can bring more people into this space? What are some barriers for new modelers?
Doyne: The field has matured. We now understand better how to build agent-based models, even though it’s not yet a rote process. People are excited about machine learning, but that only works when you have a ton of data and want to make observational predictions.
Agent-based models are good when data is sparse, and you need counterfactuals. They’re transparent, causal, and useful in a broader range of conditions. And of course, ML can be part of agent-based models too.
John: Do you think agent-based modeling could eventually become something that everyday people use to ask questions and influence policy?
Doyne: I hope so. I’d love to see open-source models that anyone can query. We’re figuring out how much to open source at Macrocosm, and how much to hold back for commercial use. But I want the tools to be out there.
John: Alright, closing question. What’s your "What the Flux?" moment? A time in history you wish we could’ve simulated different choices?
Doyne: The 2008 financial crisis. Not just that it happened, but how we responded. We bailed out the banks instead of the households. That may have contributed to political discontent and the rise of populism. If we’d simulated both options and seen that helping households worked better, maybe we’d be in a different place today.
John: That’s powerful. In modeling, we often see three core motivators: health, money, and votes. Do you see others?
Doyne: Those are the main ones. At Macrocosm, we’re focused on the first two. If we can improve wellbeing and economic outcomes, politics may follow.
John: Couldn’t agree more. In closing, any thoughts for listeners on why sustainability should be a core focus of modeling?
Doyne: In the long run, sustainability is inevitable. The question is whether we get there through foresight or pain. I worry we’ll have to suffer before we act, but if we build the tools now, we can help people navigate to a better path with less pain and more prosperity.
John: That’s a great place to end. Thanks again for coming on the podcast. We'll be talking with Dan from Macrocosm soon and releasing both episodes together to help build momentum for complexity economics and modeling big systems for impact.
Doyne: I appreciate it. Thank you for having me on the show.
John: Absolutely.