Haute Lumière · The reading

The economy never sits still long enough to be solved

Equilibrium was never a description of markets. It was a description of the mathematics that could be finished by hand.

The reading took an hour and changed one decision. That is the usual ratio, and it is enough.

NOTHING SETTLES

Stand at the edge of a farmers market on a Tuesday morning and try to find the part that is at rest. Fiona is explaining how to roast an aubergine to a customer who came for tomatoes. Two stalls along, a grower and his neighbour are having the same argument about pesticides they had last week and will have again next week. Children run the aisles. A busker plays, and three people who had not intended to stop, stop. Somewhere inside all of that a price is being set, and not one of the people setting it is solving an equation.

The price of Fiona's tomatoes is not her costs plus the market's appetite. It is also a late frost that took a third of the crop, a season in which salads became fashionable for reasons nobody can name, and the plain fact that she is a pleasure to buy from. Standard economics has a word for those three things. It calls them noise — the residue left over once the real forces have been accounted for, to be averaged out of the picture so the picture can be drawn.

The first move this book makes is to say that the residue is the subject. Not an error term, not a nuisance, not something better instruments would remove. The interplay between individuals is where the market's behaviour is actually manufactured, and the smoothness of the curve that replaces it is an artefact of the drawing, not a property of the thing drawn.

Equilibrium is not somewhere markets are heading. It is where a particular kind of mathematics can stop.

That is not an attack on supply and demand, which describe something real and describe it economically. It is an observation about what the tool was built to do. A model that finds the point where two lines cross is answering the question where does this rest, and it answers it beautifully. It cannot answer what does this do on the way, because the way was the part that got averaged. And for anyone trying to understand a crash, a boom, a bubble or a queue, the way is the whole question.

What follows is the alternative, worked out rather than asserted: markets as populations of interacting agents, connected in networks that have shape, coupled by feedback that has direction, learning as they go, and producing structure that nobody in them designed. The mathematics does not get harder. It gets different — a process to be run rather than a fixed point to be found.

THE CROSSED LINES

Take the market for apples, which is the cleanest case anyone has. Farmers will sell more as the price rises; buyers will buy less. Draw both, and the lines cross at a price where the quantity offered equals the quantity wanted. That crossing is the equilibrium price, and generations of students have been taught to find it, defend it, and treat any departure from it as a disturbance.

Look closely at what the picture assumes. It assumes every farmer and every buyer faces the same price at the same moment. It assumes the adjustment from one price to the next is either instantaneous or uninteresting. And it assumes the state of the market after a change resembles the state before it, so that the question is always which point are we at rather than what kind of thing is this now. Real markets break all three assumptions before breakfast.

The complexity alternative keeps the same ingredients and changes the verb. Each farmer i has a supply function relating price to the quantity they will sell; each buyer j has a demand function of their own, shaped by their preferences, their information, their budget. Nobody's function is the average function. Then price is not solved for — it is updated, period by period, by something as simple as this: the next price equals the current price plus a fraction of the gap between what was wanted and what was offered.

Written out, that is p at time t plus one equals p at time t, plus alpha times the difference between total demand and total supply. Alpha is how sharply the market reacts to an imbalance. That is the entire mechanism. It says prices rise when buyers outnumber goods and fall when goods outnumber buyers, which is exactly what the crossed lines said — and yet it behaves completely differently, because it is a rule that runs rather than a condition that holds.

Run it, and the same ingredients that produced a single tidy point now produce cycles, overshoot, periods of calm, sudden reversals. Set alpha high enough and the market overcorrects, then overcorrects the correction. None of that needed a villain from outside. The oscillation is a property of the updating itself, of the fact that everyone is reacting to a price that is already being changed by everyone's reaction.

This is the quiet indictment at the centre of the book. Equilibrium models can produce a crash, but only by importing one: a shock, an external event, something that arrives from beyond the edge of the model and knocks it off its point. A framework whose most consequential behaviour has to be delivered from outside is telling you where its boundary was drawn.

AGENTS, NOT AVERAGES

The most expensive simplification in economics is the representative agent: one consumer, one firm, standing in for all of them, behaving as the average behaves. It is expensive because the interesting properties of a crowd are precisely the ones that do not survive averaging. A crowd in which half the people are buying and half are selling has the same average as a crowd in which nobody is doing anything, and the two markets could not be less alike.

Replace the average with the actual population and a second variable appears immediately: who is connected to whom. Write the economy as a set of nodes — firms, households, funds, institutions — and let the connections between them carry weight. Formally this is the adjacency matrix, an N by N table where the entry in row i, column j says how strongly agent i acts on agent j, and a zero says they never meet. That table is not decoration around the model. It is a variable of the model, and changing it changes what the economy does.

Three investors make the point at a scale anyone can hold. Alice is cautious and buys the bond. Bob is adventurous, puts most of his money into a volatile technology stock, and says so. Carol was leaning towards the bond, hears Bob's enthusiasm, and puts a third of her capital into the same stock. Nothing irrational has happened. Yet Bob's private appetite for risk is now expressed in Carol's portfolio, and the market has an exposure that no individual chose.

Scale that to millions and the familiar pathologies stop looking like pathologies. Herding is what a connected population does when information is scarce and neighbours are visible. Bubbles are what happens when the connection carries the enthusiasm faster than the evidence. Crashes are the same channel running in reverse, and they run faster, because fear travels well.

A market is not a crowd of strangers. It is a graph, and the shape of the graph is half the outcome.

The structure of those graphs turns out to be regular in ways that matter. Watts and Strogatz showed in 1998 that even enormous networks tend to have short paths between any two nodes, so nothing in an economy is really far from anything else. Barabási and Albert showed the following year how hubs form: connections accrue to whoever already has them, so networks grow lopsided rather than even. Kirman's ants, from 1992, showed that a population with no individual irrationality in it at all can still swing en masse between two food sources, because each ant is partly recruited by the last one.

Then there is the smallest illustration in the book and one of the best. A city's coffee shops each set their own prices and bake their own pastries, and the network between them is made of nothing more than customers overhearing each other. Someone at one counter learns about the cold brew at another. The success of any single shop is written partly in the reputations of shops it has never dealt with — and no shop owner can see the graph they are in.

Nothing in the frame is at rest. The stillness belongs to the shutter, not to the morning.

THE LOOP ITSELF

A feedback loop is a circuit in which an output becomes an input. Two kinds exist, and almost everything in an economy is made of them. A reinforcing loop amplifies whatever it is given: confidence raises spending, spending raises profits, profits raise hiring, hiring raises confidence. A balancing loop opposes whatever it is given: rising prices dampen demand, dampened demand relieves prices. The first produces booms and collapses. The second produces the stability that equilibrium models mistook for the whole story.

Write the first loop down with three variables and it becomes something you can reason about rather than merely gesture at. Let C be consumption, I investment, K consumer confidence. The change in consumption depends on confidence and on investment. The change in investment is proportional to consumption. The change in confidence depends on consumption and on the confidence already present. Five constants set how strongly each arrow pulls. Nothing in those three lines is exotic, and nothing in them is controversial.

Run the system forward from a modest starting point and it grows, because each variable feeds the next and the next feeds back. That is the expansion, and while it is happening it looks like health, because by every measure in the model it is health. Employment rises. Profits rise. The confidence that started it is now being manufactured by the thing it started.

Now knock confidence down. Not catastrophically — a piece of unexpected news, the kind that arrives most weeks. Lower confidence reduces the change in consumption, because people become careful. Reduced consumption reduces the change in investment, because firms read the same signal. Reduced consumption and investment together drive confidence down further. The identical machinery that manufactured the expansion is now manufacturing the contraction, at the same speed, using the same arrows.

That is the finding worth carrying out of the chapter, and it is sharper than any warning about volatility. The boom and the bust are not opposites and they are not separate events. They are one structure seen at two moments, and a policy designed to protect the first is necessarily strengthening the second. An economy with fast, tightly coupled reinforcement is an economy that rises quickly and falls quickly, and those two sentences are the same sentence.

It also explains why economic argument so often goes nowhere. Inside a reinforcing loop, every variable can be truthfully described as the cause of every other. Confidence caused the spending; the spending caused the profits; the profits caused the confidence. Each account is correct, each is incomplete, and the search for the one that started it is a category error about circles.

WHY GROWTH BENDS

The simplest equation in the book is also the one that carries the most weight, and it fits on a line: the rate of change of X equals r times X times the quantity one minus X over K. X is however many there are of the thing. The parameter r is how fast it would grow with nothing in its way. K is the carrying capacity — the most the environment can hold.

The whole argument is in the bracket. When X is small relative to K, one minus X over K is close to one, the bracket does almost nothing, and growth is nearly exponential. As X climbs towards K, the bracket shrinks, and it shrinks the growth term itself. The brake is not applied from outside. It is a factor inside the engine, and it tightens in proportion to success.

Put rabbits in a meadow that can hold five hundred, start with a hundred, and give them an annual growth rate of 0.2. The first year's rate of change is 0.2 times 100 times one minus one fifth: twenty times four fifths, which is sixteen rabbits. The following year begins with a hundred and sixteen, and the bracket has already closed a little, so the next increment is smaller. Nothing has gone wrong. Nothing has been added. The population is slowing itself with its own numbers.

A second pass in the book uses a smaller meadow — a carrying capacity of fifty, a growth rate of 0.5, a starting population of ten — and the arithmetic gives four in the first year and 5.04 in the second, because at ten the bracket was still generous and at fourteen it had begun to bite. Two numbers, one going up while the other goes down, and between them the entire shape of a curve that rises, bends and flattens without any external hand on it.

Success is the most reliable constraint on success, and it is written into the same term that produced it.

The economic translations are immediate and they are not metaphorical. A new online marketplace opens with a hundred artisans, a monthly growth rate of 0.2, and local demand that can sustain around five hundred. The first month adds sixteen sellers. Later months add fewer, and any founder reading the raw count of new sellers will experience the flattening as a failure of marketing. It is not. It is the bracket, and the bracket was always going to close.

Here r stands for ambition, innovation, entrepreneurial energy, whatever it is that makes a thing spread. K stands for demand, attention, capital, regulatory room, the finite quantity of people who want this. Almost every argument about growth is really a disagreement about which of those two letters is being discussed — and the logistic curve is the reason a market can be genuinely healthy and visibly slowing at the same moment.

EXPECTATIONS THAT LEARN

Beatrice bakes sourdough with a crust that sings, and for years her instinct was sufficient. She could smell a change in humidity and adjust the starter for it. Then a competitor opened across the street selling fashionable baguettes, and her queue thinned, and her bread had not got worse. What had changed was outside her senses: the thing her customers were comparing her to. She began watching how he priced, how he marketed, what flour he bought. Then she baked focaccia, and tried croissants, and got the croissants wrong.

That is the whole of adaptive expectations, and it is a better description of economic life than perfect foresight ever managed. Nobody holds an accurate model of the future. What people hold is a previous guess and a recent surprise, and they move the first a little way towards the second. The formula is as plain as the behaviour: the next expectation equals the current expectation plus alpha times the difference between what actually happened and what was expected.

Alpha is the learning rate, somewhere between nought and one, and it is the interesting parameter. Near zero, a person is stubborn; the world can change around them and their forecast will not move. Near one, they are credulous, treating every fluctuation as the new permanent state. Neither extreme survives long in a market. Most real behaviour sits in the middle, which is why most real behaviour lags.

Work the coffee example through, because the arithmetic is the argument. A cup cost four dollars last month, so four dollars is the expectation. This month it costs five. The surprise is one dollar. With a learning rate of a half, the new expectation becomes four dollars fifty — above what was expected, below what was observed. That gap is not a mistake. It is a considered refusal to decide yet whether the rise is a blip or a trend, and it is exactly how a careful person behaves.

The same rule works on a trading desk with nothing changed but the units. A manager expects a stock at $105, the stock prints at $103, and with a learning rate of a half the revised expectation is $104. Do that across thousands of positions and millions of participants and something important follows: the market is not being predicted by its participants, it is being chased by them, and the distance of the chase is set by how quickly each one learns.

Which produces the genuine difficulty. Everyone is forecasting a quantity that everyone's forecasts help determine. The target moves because it is being aimed at. Rational expectations handled this by assuming the aim was already perfect and the target already still; adaptive expectations handle it by letting the aim lag, and accepting that the lag is where booms, overshoots and corrections live. Simon called the underlying condition bounded rationality as early as 1955 — not people being foolish, people deciding well with what they have, which is never enough.

She is not deciding what the market will do. She is deciding what she does if it does anything.

ORDER NOBODY ORDERED

A murmuration of starlings has no leader. Each bird tracks a handful of neighbours and adjusts, and out of that local arithmetic comes a shape visible from half a mile away — a shape no bird can see, intend or describe. Self-organisation is the name for order of that kind: structure produced from the bottom by rules operating locally, with nothing at the top holding a plan.

Markets do this constantly, and the ordinary ones are the best evidence. No authority decided that a corner of the internet should serve handmade goods, and yet crafters wanting an audience and buyers wanting something not mass-produced found each other, and a platform hardened around the traffic. Flea markets, farmers markets, auction sites, the informal trade of any city street — each arose from participants recognising an unmet need and responding to it, and each acquired its own etiquette, its own pricing conventions, its own understanding of what is and is not done.

The book gives the mechanism a form you can run. Let each producer have a popularity — a count of the buyers connected to them. At each step, popularity grows by a constant times the sum, over that producer's customers, of a satisfaction score: how good the goods were, how fair the price. Producers who please people accumulate connections. Accumulated connections bring more people, who are then also pleased. The result is not an even spread of trade across sellers but a small number of hubs holding a disproportionate share, arrived at without anyone conferring the position.

Raise the constant that governs how strongly reputation attracts, and concentration increases: the market tips further towards a handful of winners. Lower it, and trade stays dispersed. The degree of concentration in an industry is often read as evidence of someone's strategy or someone's failure to regulate. Sometimes it is only a parameter, and the parameter is how visible reputation is.

Nobody drew the shape of the market. Everybody drew one line each, and the shape is what the lines did.

None of which makes self-organised order stable. Blockbuster held the video rental market so completely that it was a synonym for it, and the rules that produced that dominance were not the rules that survived streaming. The same preferential attachment that builds a hub will build the next one somewhere else, and the incumbent's advantage is only an advantage while the connections still point at it.

This is where institutions enter, and the book is careful about what they are. Contracts and property rights are formal; trust and reputation are informal; both are feedback machinery. Strong property rights are a stabilising loop — they lower the risk of investing, so more investment happens, so the rights become worth more to defend. Weak ones run the circuit backwards into uncertainty and withdrawal. Institutions do not stand outside the self-organising market issuing instructions. They are part of it, subject to the same rise and the same fall, and the practical question is not whether to have them but which loops the ones you have are strengthening.

STRONGER AFTERWARDS

The final property in the living-systems set is the least intuitive and the most useful. Robustness is the ability to take a blow and be unchanged. Antifragility is something else: benefiting from the blow, ending in better condition than before it landed. Living systems do this routinely, which is why the concept was borrowed from them.

A forest fire is the standard case and it earns its place. In the short term it is destruction, plainly. In the longer term it clears accumulated deadwood, opens the canopy, releases nutrients, and lets species through that could not have reached light otherwise. The forest is not merely surviving the fire; the fire is part of how that forest stays a forest. Suppress fire long enough and you do not get a safer forest — you get a forest with more fuel in it.

Economies have the same structure, and the mechanism is not mystical. A downturn removes arrangements that were only viable in the last set of conditions. It reveals which dependencies were load-bearing, which of them had no alternative behind them, and which relationships existed only while money was cheap. That information is expensive and unavailable any other way, since the system will not tell you where it is brittle while nothing is pulling on it.

The 2008 crash reads clearly through this lens. It was not one institution's error arriving from outside the system. It was an emergent consequence of interactions between financial institutions, individual investors and government policy, each behaving in a way that was defensible in isolation, coupled together into a circuit whose behaviour none of them held. That is exactly what emergence means: a property of the arrangement, not of any component in it.

The design consequence is specific, and it is the most actionable idea in the book. If crises are endogenous, then preparation cannot consist of forecasting them accurately, because a system that is chasing its own expectations will not hold still to be forecast. Preparation consists of building in redundancy and flexibility — slack that looks like waste while nothing is happening, alternatives that look like duplication until the primary fails, and a willingness to let the structure change shape under load rather than hold rigid until it breaks.

None of that is an argument for welcoming damage, and the book does not make one. It is an argument about where effort belongs. Every hour spent refining a forecast of the next disruption is an hour not spent making the disruption survivable, and only one of those two activities has a known success rate.

MONDAY MORNING

A framework that cannot be acted on is a hobby. The book's practical chapters are built as a repeating circuit rather than a set of rules, and it has four positions: observe, forecast, act, evaluate. The fourth is the one most people skip, and skipping it converts the whole circuit into a guess with extra steps. An expectation that is never checked against what happened cannot learn, and a forecaster who does not learn is not applying the model in this book, whatever they believe about complexity.

Start by mapping the network you are actually in. Not as a diagram — as an inventory. The people whose finances are entangled with yours. The employer, and what that employer depends on. The bank, the platform, the fund, the landlord. Then the second ring: what those depend on. Most concentration risk in an ordinary life is invisible because it lives one step out, where a single employer, a single industry and a single housing market turn out to be the same bet wearing three names.

Then read the agents, which means reading their objectives rather than their statements. Every node in your network is optimising something, and almost none of them are optimising your outcome — not through malice, but because they have their own goals and their own constraints. A neighbour saving for a child's education and a fund maximising return to shareholders are both behaving correctly and neither is behaving on your behalf. Knowing what each party is actually solving for predicts their behaviour far better than knowing what they say.

For a portfolio, the shift is from optimisation to adaptation. A classic allocation is calculated once from historical returns and variances, and it is correct on the day it is computed and quietly wrong thereafter, because the correlations it rests on move. The complexity-informed version keeps the same instruments and changes the cadence: monitor continuously, rebalance dynamically, and treat the allocation as a position that is maintained rather than a decision that was made. A simple rule — add when a holding falls below a set threshold, trim when it rises above another — is mechanical, unglamorous, and does what it says when sentiment makes discretion unreliable.

The personal version of the same circuit is smaller and it works. Track spending for a month or two and categorise it honestly. Look for the trends rather than the totals, including the seasonal ones. Forecast the next few months from what the trends say and not from what you intend. Build the budget on the forecast. Then, after the period, compare the forecast with what actually happened and move it partway towards the truth — which is the adaptive expectations rule, applied to a household, with you as the learning agent.

Two cautions from the structure itself. Feedback in economic systems is frequently delayed, so the consequences of a decision often arrive long after the decision stops feeling connected to them, which is the strongest available argument for long horizons and against reacting to the most recent thing. And diversification, read through networks, means more than holding several assets. It means having more than one kind of skill, more than one source of income, and more than one community — because a portfolio of correlated holdings is one holding, and so is a life.

THE LIVING FIGURE

Two pictures of an economy have been in competition throughout. In the first it is a machine: components with fixed relations, inputs producing proportional outputs, a designer somewhere, a correct setting to be found, malfunction when it departs from that setting. In the second it is something closer to an organism — flows and stocks, circuits that amplify and circuits that damp, parts that adapt because they are alive, and a shape that persists while every element of it turns over.

The machine picture is not stupid and it was not a waste. It produced national accounts, monetary policy, the whole apparatus by which a modern state can see itself, and none of that is being discarded here. What the book argues is narrower and harder to dismiss: the machine picture is a special case, accurate over short intervals and small departures, and the events people most want explained all live outside the interval where it holds.

Read the organism picture honestly and prediction has to be demoted. Not abandoned — demoted, from the goal to a tool among others. In a system where agents learn, where learning changes the thing being learned about, and where structure emerges from interactions nobody supervises, a confident forecast of a specific future is a claim about a system that does not have one. What can be known is different in kind: which loops are present, which direction they run, where the couplings are tight, what the system does when it is wrong.

The useful question was never what the economy will do. It is what this economy does when something it did not expect arrives.

There is a plainer consequence for anyone who has been made to feel stupid by economics. If you have read the textbook account of rational agents converging on equilibrium and found it did not describe any market you have stood in, you were not failing to understand the model. You were noticing, correctly, that the model had set aside the part you were looking at. The instinct was sound. What was missing was the framework that takes it seriously, and the framework exists, and it has arithmetic in it.

That is what the volume is: thirteen chapters and 48,418 words that take the argument from equilibrium through networks, feedback, learning and self-organisation, with the mathematics written out step by step and the numbers worked rather than gestured at. It sits on the Living Systems Economics shelf at Haute Lumière, and every chapter of it is free to read, in full, without an account. Buying a copy is for keeping the file — the EPUB, the PDF, the press edition. The reading is open either way, and it is open now.

Free to read

Free to read, and free to hear. Every chapter of every book in this house, and every narration of it, is open to anybody. No account, no card, nothing to cancel.

What is in it

• From Equilibrium to Emergence: Reframing Economic Thinkingopen this in the house search
• Networks and Agents: The Building Blocks of Complex Economiesopen this in the house search
• Feedback Loops and Nonlinear Dynamics: Understanding Economic Fluctuationsopen this in the house search
• Adaptive Expectations and Learning: How Markets Evolve Over Timeopen this in the house search
• Self-Organization and Market Structure: The Rise and Fall of Institutionsopen this in the house search

A model that must import its own crashes has drawn its boundary in the wrong place.

The representative agent is an average of people, none of whom exist.

Prediction asks what happens next. Preparation asks what this does when it is wrong.

Every reinforcing loop is a boom read forwards and a bust read backwards.

Keep looking

Every phrase on this page opens into the house search. The shelf holds Living Systems Economics and six other shelves, and the reading is free.