# Making Sense of Chaos **J. Doyne Farmer** | [[Prediction]]  --- > "By definition, a system is complex if it has emergent properties." Standard economics assumes rational agents optimising in equilibrium. The economy only changes when external shocks disturb it. But real economies exhibit endogenous motion—they never settle down, even without outside disturbances. Real agents aren't rational optimisers; they're boundedly rational, using simple heuristics to navigate uncertainty. And real economies are complex adaptive systems with emergent behaviour that cannot be predicted from individual components alone. J. Doyne Farmer's argument: we need complexity economics to understand how economies actually work. Standard models use equations solved by hand or computer, assuming homogeneous rational agents. Complexity economics uses agent-based simulations with heterogeneous, boundedly rational agents. The difference matters because complicated, competitive systems don't converge to equilibrium—they exhibit chaotic dynamics. Equilibrium models can't explain business cycles, financial crises, or technological change. Complex systems can. The operational insight: the economy is an ecosystem of specialists organised as networks. Production networks determine innovation rates through trophic levels. Financial systems evolve like biological ecosystems through variation and selection. Nonlinearity creates emergence, and simulation is often the only method for understanding it. Better models require better data, bottom-up construction, and accepting that the whole is different from the sum of its parts. --- ## Core Ideas ### [[Complexity Economics]] Complexity economics is an interdisciplinary movement using principles completely different from standard economics. Standard economics assumes forward-looking agents intelligent enough to reason through any problem. Complexity economics assumes boundedly rational agents who make imperfect decisions with limited reasoning ability. Standard economists write models as equations and solve them by hand or computer. Complexity economists use agent-based models—computer simulations that can model the diversity and heterogeneity of real actors. The mathematical tools come from a much broader palette: dynamical systems, statistical physics, ecology, evolutionary biology. The principle of verisimilitude: models should fit the facts and their assumptions should be plausible. Don't assume equilibrium and rational expectations just because the maths is easier. Model what's real. ### [[Complex Adaptive Systems]] A system is complex if it has emergent properties—behaviour qualitatively different from what any building block can do alone. Complex adaptive systems like brains, ant colonies, and economies are distinguished by the fact that their properties have evolved over time through selection. Chaos is characterised by two essential properties: sensitive dependence on initial conditions, and endogenous motion—even without external shocks, the system never settles down to rest. Standard macroeconomic models have only fixed-point attractors. If left alone, the model economy settles into rest and stays there forever. All changes must be driven by exogenous shocks. This rules out endogenous business cycles from the outset. Chaos can be simple or complicated. Simple chaos (like the Lorenz equations) requires only a few variables. Complicated chaos (like weather) requires many variables acting independently—many degrees of freedom. The economy likely sits somewhere on this spectrum, which is why simulation is essential. ### [[The Economy as Ecosystem]] The economy is an ecosystem of specialists. In biology, an ecosystem refers to a collection of species who interact with and affect each other. Each is a specialist with its own unique strategy for extracting energy from the environment. Species eat each other, compete, cooperate, and collectively alter the environment. To understand grass, you must think about lions—lions protect grass by controlling zebras. Similarly, the economy is a complex ecosystem of specialised organisations populated by workers who belong to households. Understanding this ecosystem means thinking in terms of networks, which provide a universal language for describing complex systems. The skeleton of the modern economy is a vast network of balance sheets. Accounting is represented by the network of balance sheets, which continually changes as people make economic decisions. Accounting is complicated but well-understood; decision-making requires an understanding of human nature that is still incomplete. ### [[Production Networks and Trophic Levels]] A production network transforms natural resources and labour into goods and services for consumers. It's the economy's metabolism—a network of firms analogous to the network of chemical reactions in living cells. **The typical industry improves more due to the effort of its suppliers than due to its own effort.** Roughly 65% of improvements in a given industry come from other industries—only 35% come from within the industry itself. Industries with higher trophic levels (deeper supply chains) improve faster because they have more industries below them providing innovations. This helps explain why manufacturing industries tend to improve faster than service industries. The predictions don't depend on human behaviour directly—trophic level is a structural property of the economy. Specialisation doesn't just make the economy work better; it automatically boosts the rate of innovation. ### [[Bounded Rationality and Heuristics]] In 1955, Herbert Simon introduced bounded rationality: when confronted with hard problems, cognitive limitations make perfect rationality impossible. We seek "good enough" choices rather than optimal ones. Real people don't solve day-to-day problems by doing complicated maths calculations. Heuristics are simple mental processes that humans, animals, and organisations use to quickly form judgments, make decisions, and find solutions to complex problems. Psychology experiments confirm that real people rely on them. In uncertain environments, simplicity is a virtue—there are many situations where heuristics make better predictions than complicated methods. This is the less-is-more effect. Following Simon's footsteps, Gerd Gigerenzer and economist Reinhard Selten proposed that we draw on an "adaptive toolbox" of heuristics. When testing models, focus on their ability to make predictions about the future, not their ability to fit past data. ### [[Networks as Universal Language]] Networks are one of the core ideas in complex systems. They identify essential building blocks and supply a schematic view of their interactions. There are usually many possible choices for nodes and links, depending on what you want to understand. Representing a system as a network begins by identifying building blocks and interactions: What are the most important nodes? Are there communities—sets of nodes that interact with each other much more strongly than with other nodes? Network modelling is one of the few complex-systems ideas that has made its way into mainstream economics. Contracts form a vast web connecting us in space and time. Each contract sits on someone's balance sheet as an asset or liability. Credit connects us by linking the balance sheets of borrowers and lenders. The network of credit contracts is an important subset of the web of contracts. The strongly nonlinear nature of leverage is a powerful amplifier that can cause booms and busts. --- ## Key Insights **Complicated means many moving parts; complex means emergent behaviour.** The words aren't synonyms. The Lorenz equations have only three degrees of freedom but create complicated geometric objects with complicated motion. The attractor is complicated but its underlying cause is simple. Complicated systems can have simple underlying causes. **Nonlinearity is a necessary condition for emergence.** Without nonlinearity, there would be no emergence and no complex systems. A system is nonlinear if the whole is different from the sum of its parts. In contrast, a linear system has a whole equal to the sum of its parts. In a world where almost everything is nonlinear, computer simulations are essential. **Equilibrium is the wrong assumption for complicated, competitive settings.** Games converge to static equilibrium if they're non-competitive or simple. If they're both competitive and complicated, they're unlikely to converge—instead, their dynamics will be chaotic. Standard economic models in any complicated, competitive setting are very likely wrong. **Financial markets are ecosystems that evolve.** The fact that investors are specialists makes ecology the natural framework. The financial ecosystem satisfies the essential elements of evolution: descent with variation and selection. Descent means continuity across generations, variation means offspring differ from parents, selection means only successful individuals survive and reproduce. **Economic forecasting lags far behind weather forecasting.** Weather forecasts have systematically improved over time. Economic forecasting has never been very accurate and has made only slow progress over eighty years. Why? Weather is modelled from the bottom up at the finest scale possible, whilst government economic models operate at the aggregate level with a single representative household. Weather models are highly nonlinear and closed (covering the entire world). Economic models often use linear approximations and are open (modeling single countries, treating outside influences as exogenous shocks). **Wright's Law and Moore's Law predict technological change.** In 1936, Theodore Wright observed that whenever cumulative production of a given type of airplane doubled, its cost dropped by about 20%. This applies to many technologies, though the rate varies substantially (and in most cases is close to zero). Because of wide variation in improvement rates, supporting some technologies is far more effective than supporting others. These laws provide the best tools we have for predicting future performance of a technology. **Data is the principal driver of scientific progress.** There's bidirectional feedback between data and theory: data inspires development of theories and makes it possible to test them, whilst theories determine what data we collect. Theories can only be proved wrong or right by data. When this doesn't happen, theories become detached from reality and science gets lost. Better data is central to improving agent-based models. --- ## Connects To - [[Linked]] - Both focus on networks as the fundamental language for understanding complex systems - [[Antifragile]] - Taleb on systems that gain from disorder connects to Farmer on nonlinearity and emergence - [[Black Box Thinking]] - Both emphasise learning from data and testing predictions rather than fitting past data - [[Algorithms to Live By]] - Computer science heuristics complement Farmer's bounded rationality and adaptive toolbox - [[The Fifth Discipline]] - Senge on systems thinking aligns with Farmer's view of the economy as a complex adaptive system - [[Prediction Machines]] - AI economics complements complexity economics on prediction under uncertainty --- ## Final Thought Standard economics assumes equilibrium because the maths is tractable. Rational agents optimise, markets clear, and the economy settles into rest unless disturbed by exogenous shocks. But real economies exhibit endogenous motion—business cycles, financial crises, technological revolutions—without external disturbances. Real agents use simple heuristics under uncertainty, not complicated optimisation. Real systems are nonlinear, and the whole is different from the sum of its parts. Farmer's complexity economics replaces equations with simulations, homogeneity with heterogeneity, equilibrium with chaos. Agent-based models can capture what standard models rule out by assumption: endogenous change, emergent behaviour, networks of specialists that evolve through variation and selection. The economy as ecosystem changes how you think about prediction. Production networks determine innovation rates through trophic levels—industries improve faster when they have deeper supply chains. Financial markets evolve like biological ecosystems—investors are specialists, strategies are phenotypes, and only successful variants survive. Leverage creates nonlinear amplification that drives booms and busts. The gap between economic and weather forecasting shows what's missing. Weather models are bottom-up, highly nonlinear, global, and continuously tested against reality. Economic models are top-down, often linearised, nation-specific, and rarely validated on out-of-sample predictions. Better models require better data, agent-based construction, and accepting that complicated competitive systems don't converge to equilibrium—they exhibit chaotic dynamics. You can't understand the economy by studying isolated components. Everything depends on everything else. Networks provide the language. Emergence requires simulation. Nonlinearity makes the whole different from the sum of its parts. That's what complexity economics captures that standard economics misses.