Chaos and Complexity Economics (with J. Doyne Farmer)

Episode Summary

Executive Summary: Russ Roberts interviews physicist Dowen Farmer about complexity economics: using agent-based, simulation-based models rather than representative-agent optimization to better understand and predict messy, real-world systems. They discuss housing, COVID, weather forecasting, and why richer data and heterogeneity can improve policy analysis and ex-ante prediction, while mainstream economics remains influential but limited in dynamic crises.

Main Topics: What complexity economics is (Priority: 5/5): Farmer defines complexity economics as applying complex-systems science to economics, emphasizing simulation, heterogeneity, and dynamic interactions over utility-maximization-based equations. Why agent-based models can outperform standard macro models (Priority: 5/5): The discussion contrasts top-down equilibrium models with bottom-up simulations that allow heterogeneous agents, institutional detail, and disequilibrium dynamics. Housing markets and price formation (Priority: 5/5): Farmer uses housing to show how real prices are set by aspiration, comparables, and gradual markdowns, not instant market clearing, and how lending standards fueled the 2008 bubble. Prediction, counterfactuals, and policy use (Priority: 4/5): Roberts and Farmer debate whether economists need theory versus data, with Farmer arguing causal simulation is essential for counterfactual policy analysis when historical analogs are scarce. Weather forecasting as a template (Priority: 4/5): Farmer compares economics to weather prediction: big gains came from physics-based simulation, heavy investment, and localized data, suggesting economics could improve similarly. COVID modeling and universality (Priority: 4/5): Farmer cites his COVID model as a proof of concept, arguing that salient structural features and universality can make some economic shocks predictable even amid complexity. Professional resistance and diffusion (Priority: 3/5): The conversation ends on why complexity economics has not penetrated mainstream economics—career incentives, jargon, elegance of standard models, and stronger acceptance in central banks and commercial settings.

Key Arguments: Complexity economics replaces utility-maximization-first modeling with simulation of individual agents and institutions, which is better suited to messy problems like macroeconomics and climate change. Agent-based models can incorporate heterogeneous behavior, learning, geography, race, income, and other real-world features that standard models abstract away or cannot tractably solve for. The key advantage is not merely more realistic description, but better ex-ante prediction and more credible counterfactual policy analysis when the future differs from past data. Housing prices often do not clear instantly; they adjust through aspiration pricing and repeated markdowns, producing sluggish dynamics and large buyer-seller imbalances. The 2008 housing bubble was driven more by changes in bank lending standards than by interest rates alone, something detailed simulations could capture because they include actual loan types and borrower characteristics. Behavioral realism and macro forecasting are often disconnected in mainstream economics; complexity economics tries to bridge that gap by embedding human behavior directly into macro models. Weather forecasting improved only after meteorology moved from statistical analogy to physics-based simulation; Farmer argues economics can follow the same path with sufficient investment and data. COVID showed that some macro shocks are predictable if models focus on salient flow constraints such as labor availability, inputs, demand, and industry interdependence. Mainstream economics is not useless—supply and demand still matter—but its workhorse models are too rigid for many out-of-equilibrium problems and counterfactual policy questions. Adoption lags because economics is an entrenched profession with strong career incentives and a taste for elegant, tractable mathematics, while complexity models are newer and harder to market.

Data Points: Episode date: August 1st, 2024 - Introductory metadata from the host Housing data horizon: a decade and a half - Farmer says his Washington, D.C. housing analysis used 15 years of transaction data Agents in simulations: a million agents - Farmer describes running large agent-based models with many individual decision-makers Variables in standard models: more than a dozen independent variables - He cites mainstream economists saying standard models become unsolvable beyond this point COVID GDP forecast, UK Q2 2020: 21.5% predicted decline - Farmer cites his model’s real-time prediction Actual UK GDP decline, Q2 2020: 22.1% - Reported outcome compared with the model prediction Forecast error comparison: off by a factor of 20% - Host notes the Fed’s model was much less accurate when simulating a 20% housing-price drop Weather forecasting improvement: one extra day of accuracy per decade - Farmer says forecast skill improved by about a day every ten years Weather forecast development period: 1950 to 1980 - He describes the major effort to build physics-based weather models Weather forecasting investment: billions of dollars - Large public investment enabled modern numerical weather prediction Occupation-level data: about 500 occupations - Used in the COVID model to estimate workplace proximity and exposure Loan mix in housing bubble: vanilla 30-year fixed loans vs. balloon payments and smaller down payments - Farmer contrasts old lending norms with looser pre-crisis lending practices Model scale in roulette project: first wearable digital computer - He and colleagues built custom hardware to predict roulette outcomes

Pivotal Quotes: "It means doing economics in a different way than mainstream economists do it." — Dowen Farmer: Farmer defines complexity economics at the start of the interview "We’re not limited by complication. If people behave in a more complicated way, fine. We can write a computer program that mimics that." — Dowen Farmer: On why simulation can handle heterogeneity and complexity better than closed-form equations "What we’re interested in is how the herd behaves typically rather than how a few isolated individuals behave." — Dowen Farmer: On the goal of agent-based modeling and macroeconomic realism

Implications: The interview suggests policy institutions should pair richer data with simulation-based models for crises, housing, climate, and public health. Complexity economics may gain influence first in central banks and commercial forecasting before academia fully accepts it.