What this covers

Doyne Farmer, a physicist, argues for agent-based modeling as an alternative to mainstream economic theory. Rather than solving equations based on utility maximization and rational agents, Agent-Based Modeling simulates millions of heterogeneous, rule-following agents interacting in a computational economy. Farmer makes the case to Russ Roberts that this approach produces more accurate predictions, especially when systems become complicated, and better captures how humans actually behave. The conversation spans the theoretical foundations of this shift, the empirical track record of agent-based models, and the institutional resistance within academic economics to adopting them.

The discussion covers several concrete domains. Farmer walks through how equation-based models become mathematically intractable once they contain more than roughly a dozen independent variables, whereas agent-based simulations can handle a million agents because realistic decision rules make them tractable. He examines Salient Features Predictability—the insight that certain structural features can render an economy predictable even without knowing all individual behavioral details—and contrasts this with Hayek's worry about dispersed knowledge. The COVID-19 pandemic serves as a live case: Farmer's model predicted a 21.5% UK GDP decline for Q2 2020 versus an actual 22.1%, and he explains how supply-chain topology and occupational proximity, modeled at the agent level, allowed that accuracy. Housing markets illustrate a mechanism called Aspiration-Level Adaptation, where prices adjust through seller behavior rather than instantaneous market clearing. Farmer also diagnoses why central banks are adopting agent-based approaches while academic economics remains locked into workhorse models that embed rational expectations despite behavioral evidence to the contrary, and he draws a parallel to weather forecasting, which stalled for a century before computational physics transformed it.

Sharpest takeaway

Farmer argues that complexity economics—simulating economies as systems of millions of heterogeneous, rule-following agents rather than solving utility-maximization equations—produces more accurate, detail-rich, ex-ante predictions than mainstream economics, especially when problems get complicated.

  • Agent-based simulations can absorb arbitrary real-world complexity that equation-based models cannot solve once variables exceed about a dozen
  • The COVID model made an accurate ex-ante GDP prediction (21.5% predicted vs 22.1% actual UK Q2 2020 decline)
  • Salient structural features (like supply-chain flows) can make systems predictable even without knowing all individual behavioral details

The argument · threads6 threads · 27 claims
0.75

Agent-based models incorporate full heterogeneity and avoid intractability by simulating realistic populations rather than deriving solutions from stylized equations.

4 pointscentrality 5/5
  • Agent-based models can incorporate full heterogeneity—income, race, gender, geography, age, education—by building synthetic populations of a million individuals matched to census data, whereas mainstream macro must use stylized distributions of infinitesimal agents that can only capture one feature (typically income) at a time, which matters because poor and wealthy people behave very differently.

    “In our models, we can put in income, race, gender, geography. Whatever you want, we can put it in there. And we... do build synthetic populations in our macro models. We might have a million individuals in our population. Those individuals are chosen using census data to match real people.”

  • Because agent-based models are intrinsically dynamic and model the world in a disequilibrium way, equilibrium (if it occurs) is an emergent phenomenon the model produces, giving these models the capacity to get the dynamics of the path from one state to another right rather than merely asserting the system eventually lands at equilibrium.

    “If an equilibrium happens, it's an emergent phenomenon. It's something that the model does... because we're modeling in a disequilibrium way and we're actually explicitly making a dynamical model, we have the capacity to get the dynamics right, to get the path from A to B rather than just saying, eventually we're going to land on B.”

  • First-principles equation-based economic models become intractable once they contain more than roughly a dozen independent variables (solution times measured in centuries), whereas agent-based simulations can run with a million autonomous agents because realistic decision rules make the models simpler to run, not harder.

    “Once you have more than, say, a dozen independent variables in a standard model, you can't solve it anymore. The solution times are measured in centuries.' In contrast, we have models with a million agents in them all acting of their own volition.”

  • Complexity economics applies complex-systems science to economics by simulating the economy with computer programs rather than using utility maximization and equation-solving to derive what people will do.

    “it's the applications of complex systems, science, and methods, to economics... It means simulating the economy rather than using utility maximization to write down equations to solve for what people will do.”

0.75

Mainstream economics lacks causal first-principles structure needed to evaluate novel policies and refuses to incorporate known behavioral realities into workhorse models.

4 pointscentrality 5/5
  • Federal Reserve economists were worried about falling housing prices before 2008 and asked their best model (FRB/US) what would happen if housing prices dropped 20%; the model said 'no big deal,' and in hindsight the model was off by a factor of 20—so it was the model, not the economists' attention, that failed.

    “they asked FRB/US, a Federal Reserve Bank US model, their best model, what happens if housing prices drop by 20%? And then, said, 'Oh, not much. No big deal.' And, as they themselves said, with hindsight, the model was off by a factor of 20.”

  • Economics is at a crossroads where it is confused: any sensible economist acknowledges people are not rational, but the field does not yet know how to put real behavior into the workhorse models used to make policy.

    “my view is economics is at a crossroads where it's confused. Where, any sensible economist has acknowledged that people are not rational, but they don't really know yet how to put that into the models”

  • Causal first-principles models (not just econometrics) are essential for evaluating counterfactual policy changes, because econometricians can only infer the effects of a policy if historical natural experiments exist; when a policy enters a never-before-seen world, only models with built-in causal relationships can make predictions.

    “if you want to consider counterfactual situations, you really have to understand cause/effect because, when you make a policy change, you're going into a world that may be different than any world you've ever seen before.”

  • There is cognitive dissonance in economics between behavioral economists who describe how people really behave and the macro-modelers at central banks who still build policy models based on rational expectations plus 'frictions,' because the field acknowledges people are not rational but cannot mathematically incorporate realistic behavior into its workhorse models.

    “there's cognitive dissonance between the behavioral economists--who are saying this is the way people behave in economic settings--and the macro-modelers who build models for places like the Federal Reserve or the U.S. Treasury... those two things are at odds because right now the models for places like the Federal Reserve are still based essentially on rational expectations.”

0.61

Agent-based models have demonstrated real predictive accuracy on major events like COVID and housing, validating the approach empirically against mainstream failures.

4 pointscentrality 4/5
  • The 2008 housing bubble was primarily caused by a shift in bank lending policy—from vanilla 30-year fixed-rate 20%-down loans toward complicated loans with balloon payments and smaller down payments—rather than by interest rates, which had only a small effect, as shown by simulating the actual loan characteristics in the Washington D.C. data.

    “The housing bubble was caused by a shift in lending policy by banks. Banks got a lot looser in who they were giving loans to... we could see that these much more complicated loan types, which were much looser, was a thing that fueled the bubble.”

  • Farmer's complexity-economics COVID model made a real-time ex-ante prediction of a 21.5% decline in UK GDP for Q2 2020, against an actual outcome of 22.1%, demonstrating that agent-based models can make accurate forward-looking predictions.

    “We predicted a 21.5% decline in GDP [Gross Domestic Product] in the United Kingdom in second quarter of 2020. The answer was 22.1%.”

  • The COVID model worked with minimal behavioral assumptions—essentially that a company cannot produce without inputs, labor, and demand—by mapping how close people work in ~500 occupations, which industries use which occupations, and then iterating daily to watch inventories run out and supply shocks propagate downstream colliding with demand shocks propagating upstream.

    “The only assumptions we make in there are that a company can't produce its good if it doesn't have the inputs, if it doesn't have the labor, or if it doesn't have the demand... We could see the supply shocks propagating downstream, and the demand shocks propagating upstream. We could see them colliding. And that's why the model worked well.”

  • Housing prices are not set by instantaneous market clearing but by 'aspiration-level adaptation': sellers list at a hoped-for comparable-based price and progressively mark it down until it sells or they refuse to go lower, which makes prices react sluggishly and allows the market to be far from clearing—evidenced by a decade-and-a-half of Washington D.C. sales data.

    “They get set by what's called 'Aspiration-Level Adaptation.'... If it doesn't sell after a month or two, they mark it down... it means that prices react very sluggishly to changes in the housing market. It also means that the market can be far from clearing.”

0.56

Economics is a closed profession that discourages alternatives and blocks careers, whereas central banks pragmatically adopt agent-based models when their reputation is at stake.

3 pointscentrality 4/5
  • Several central banks (Bank of Canada, Bank of Italy) are starting to use agent-based models because their reputation is on the line when they make bad predictions, making them less dogmatic and more open-minded than economics departments, which is Farmer's strategic channel to pressure academia.

    “Bank of Canada is starting to run an agent-based macro-model. Bank of Italy. Several Central Banks are starting to use these ideas because their reputation is on the line when they make bad predictions; and there's less dogmatism.”

  • Economics is a fairly closed profession in which an advisor would warn a graduate student that building agent-based models is a bad career choice that would block jobs at top departments, because the field has locked into a single point of view and is unwilling to admit competing approaches.

    “your advisor will say, 'Sorry. That is really a very bad career choice. Because if you do that, you're never going to get a job at Harvard or any other American university in an economics department.'”

  • Farmer states his book is not a critique of mainstream economics; he removed criticizing material and focuses on presenting the alternative, asking only that the mainstream allow competing ideas to be tested against data.

    “my book is really not a critique of mainstream economics. I actually removed all the stuff criticizing... my main critique would just be: Allow more room to let these other ideas in and let them compete against your ideas.”

0.44

Randomness and complexity are subjective features depending on information and models; economies may be simpler than weather because some features remain predictable.

6 pointscentrality 2/5
  • Farmer and Norman Packard, with a team of about 20 people, beat roulette by building the first wearable digital computer (hidden under an armpit with battery packs and toe switches) that predicted roughly where the ball would land 6-10 seconds ahead, giving them an edge over the house.

    “we built the first wearable digital computer, which was concealed under an armpit with a pack of 12 AA batteries... The computer would make a prediction about roughly where the ball was likely to land, about 6-10 seconds ahead of when the ball actually landed.”

  • Contrary to the intuition that people thinking makes economics harder than weather, the economy is in many ways simpler than the weather—requiring less computing—because weather is fundamentally chaotic, whereas salient features can make the economy predictable even without knowing all the individual details Hayek worried about.

    “The weather is really complicated. The economy is actually simpler. The weather is fundamentally chaotic... But, I think the key point is that there can be salient features that allow predictability even when we don't know all the details that Hayek is worried about.”

  • The physics concept of universality—where, within a given universality class, microscopic details don't matter and behavior is more or less the same—applies to macroeconomic models, as mapped by Jean-Philippe Bouchaud's group, meaning you don't need every detail right, only certain criteria, until you cross a boundary into a different class.

    “This relates to something in physics called universality, which is: It's well known in physics that there's many situations... where the details don't matter. As long as you're in what's called the right universality class... the behavior is more or less the same.”

  • Randomness is subjective: something can appear random and unpredictable with one set of information but become predictable with different information—as with the roulette ball, which becomes predictable once you measure its position and velocity and know the forces acting on it—and the same may hold for economics with better models and richer data.

    “randomness is a subjective thing. Something can look random and unpredictable with one set of information; but if you have different information... then that thing that was previously random is no longer random.”

  • Supply and demand is a genuine and major force underpinning economics that complexity economics retains in all its models; the distinctive edge is the ability to model out-of-equilibrium situations where supply does not match demand and what happens on the way to equilibration.

    “all the models that I discuss in the book have supply and demand in them in one way or another... our edge there is that we can really talk about out-of-equilibrium situations where supply doesn't match demand and what happens on the way to supply equilibrating with demand.”

  • Larry Summers, a mainstream economist, read Farmer's book, provided substantive feedback, and endorsed it—an example that a few open-minded mainstream economists engage with complexity economics.

    “I was tickled that Larry Summers actually endorsed my book... to my utter surprise, actually read the book and provided substantive feedback.”

0.38

Weather forecasting improved by simulating fundamental physics on computers after stagnating under statistical methods, offering a historical roadmap for economics.

2 pointscentrality 3/5
  • Weather forecasting accuracy stagnated for ~100 years (late 19th century to 1980) under statistical analogy methods, but improved dramatically after 1980 when forecasters began simulating the fundamental physics of weather on computers—an effort requiring billions of dollars from 1950 to 1980—offering a model for what economics could become.

    “for roughly a hundred years, from the late 19th century to about 1980, the accuracy of weather forecasting stayed about the same... Predictions of the weather got better starting in 1980 because they used a fundamental model.”

  • Weather forecasting accuracy improves such that roughly every decade, forecasts for one additional day into the future become as accurate as forecasts for the prior day were a decade earlier (e.g., a two-day forecast a decade ago equals a three-day forecast now).

    “if we could forecast with a certain level of accuracy two days ahead, a decade ago, we can now forecast with that same accuracy three days ahead.”