An alternative theory of economic cycles
An alternative theory of economic cycles
The Local-Maxima Theory of Boom and Bust
Why economies crash when they stop climbing, not when they climb too far.
The puzzle
The crashes that built the modern world
The technologies that crash hardest are often the ones that end up mattering most. British railways drew a speculative mania in the 1840s, then the share prices collapsed and ruined thousands of investors. Within a generation, railways had rebuilt the geography of trade, labor, and even the way clocks were set. The internet repeated the trick. In 2000 the dot-com index gave back years of gains in a few months, and the obituaries were brutal. A decade later the same technology ran retail, advertising, media, logistics, and the social lives of billions.
If busts were nothing more than markets correctly punishing bad ideas, this would be a strange coincidence to keep meeting. The market punishes the idea, and then the idea wins anyway. Something in the standard account is missing, and it is missing in a consistent direction.
The orthodox view
What the excess story gets right, and what it leaves out
The orthodox explanation is familiar and partly correct. A promising technology appears. Expectations run ahead of reality. Capital pours in. Prices climb past anything the fundamentals can justify. Then reality disappoints, confidence breaks, and the correction is swift and painful. Hyman Minsky added a darker turn to this picture: calm itself breeds the next crisis, because long stretches of stability tempt investors to borrow more and hedge less, until the structure is one shock away from failure. On this reading, stability is dangerous because it makes people take on more financial risk than they notice.
That story explains a great deal. It does not explain the pattern we started with. If the dot-com crash were only the market discovering that internet companies were worthless, the internet would have stayed worthless. It did the opposite. The excess-and-correction account is sharp about the money. It is quiet about the thing the money was chasing, which is real technical and institutional progress. And it has almost nothing to say about why so many busts are followed, after a pause, by exactly the breakthroughs the bust seemed to rule out.
The claim
Stopping too soon
Here is the alternative, stated plainly. Many busts happen not because a system went too far, but because it stopped too soon. Success makes a system cautious. Caution makes it polish what already works instead of searching for what might work better. The system settles onto a comfortable peak and defends it. Real progress slows while expectations, formed during the fast years, keep climbing. The gap between the two widens until it snaps. That snap is the bust.
The cause is not runaway acceleration. It is premature convergence.
Systems rarely collapse because they keep climbing. They collapse because they stop.
The mental model
Peaks, valleys, and why the best move is sometimes backward
The cleanest way to see this comes from evolutionary biology. In 1932 Sewall Wright drew what he called an adaptive landscape: a terrain of hills and valleys where height stands for fitness, for how well a design suits its environment. A population climbs the nearest hill through small improvements, each step a little better than the last. Eventually it reaches a summit. From there, every direction is downhill, so it stops.
The trouble is that the summit it reached may be a modest one. A taller mountain can stand close by, separated by a valley. To reach the higher ground, the population would first have to go down, to get worse before it could get better, and ordinary improvement never chooses to get worse. So it stays on the smaller hill. Biologists call this being trapped on a local maximum.
Economies behave the same way, for the same reason. A firm, an industry, or a whole innovation ecosystem climbs toward a successful design and learns to run it extremely well. The better it gets, the more every small move away from the winning formula looks like a step down. The skills, the assets, the supplier relationships, the customer habits, the regulations, all of it hardens around the current peak. The higher mountain is still out there. Reaching it would mean a descent nobody wants to take. The very success of the design becomes the reason the system cannot leave it.
This is the core of the theory: success is the mechanism that prevents the next success.
The mechanism
How a winning system talks itself into standing still
There is a well-known tension in how any organization spends its effort. James March named the two sides in 1991: exploration and exploitation. Exploration is the search for new approaches, with uncertain and often distant payoffs. Exploitation is the refinement of what already works, with payoffs that are smaller but reliable and soon. Healthy systems do both. The problem is that success quietly tilts the balance toward exploitation, and several forces push in the same direction at once.
Start with fear. Once there is a lot to lose, the people in charge worry more about protecting the gain than chasing the next one. Investors ask for predictable returns. Boards reward the manager who defends the franchise over the one who would bet it. Regulators, having watched the disruption the technology already caused, lean toward stability. Each of these is reasonable on its own. Together they raise the cost of exploration and lower the cost of standing pat.
There is a clean way to describe what this does. Fear shortens time horizons, and a shorter horizon mechanically favors exploitation. Exploration pays off slowly and unevenly, with most of the value sitting in a few far-off, hard-to-predict wins. Exploitation pays soon and steadily. The moment a system starts discounting the future more heavily, whether from caution or from a fear that a bust is coming, it shifts effort away from search and toward refinement, without anyone announcing a decision to stop innovating. The retreat from exploration is just the sum of many sensible short-horizon choices.
Now watch what this does to expectations. During the boom, real capability rose fast because the system was exploring, and people set their expectations by extrapolating that pace. Once exploration gives way to refinement, real progress decelerates. You are polishing, not discovering. Expectations, anchored to the boom-era rate and carried by momentum, keep climbing for a while after the underlying progress has bent toward a plateau. The gap between what people expect and what the system actually delivers widens quietly, then gets recognized all at once. The recognition is the bust.
Notice the loop hiding in here. The fear of a coming bust shortens horizons; shorter horizons starve exploration; starved exploration produces the stagnation that brings on the bust. The expectation of a crash can help cause the crash.
Two engines, one collapse
Why Minsky and this theory are both right
It helps to separate two kinds of bust that usually arrive tangled together.
The first is the excess bust. The system identifies a real peak, and capital overshoots it. The asset is worth something; the price is wrong. This is the classic story, and Minsky describes its financial plumbing well.
The second is the convergence bust. The system climbs a real peak, then quits searching. Expectations extrapolate a climb that has already stopped. Here the current peak may even be priced fairly. What fails to arrive is the higher peak the boom implicitly promised, because the search for it was abandoned. The disappointment is not that the system was worth less than people thought. It is that the system stopped becoming worth more.
Real episodes blend the two. The contribution of this theory is to make the convergence component visible, because it hides in a place the usual models do not watch: in research budgets, in the narrowing range of approaches anyone is willing to fund, in the slow hardening of an industry around a single design.
This is also why Minsky and the local-maxima view are not competitors. They describe two halves of the same trap, and they fit together with an odd symmetry. Minsky's agents grow more financially adventurous as calm sets in; they borrow and bet more. The agents in this theory grow more technically conservative as success sets in; they search less. Both can run at the same time, and when they do you get the most combustible setup of all: money getting bolder while ideas get timider. Financial risk-taking rises while real exploration stalls. The bust is the moment the money finally notices that the ideas stopped moving.
The record
Reading the old crashes again
Go back to the famous busts and ask a different question. Not "how crazy were the valuations," but "what happened to exploration."
The dot-com crash is the clearest case. The internet did not fail in 2000. The ecosystem was caught between an early web of cheap experiments and a mature digital economy with real ways to make money, and the second one was not built yet. When the first business models broke, capital fled before the transition finished. What makes the story end well is what survived the crash: a flood of cheap engineering talent looking for work, a build-out of fiber and data centers that did not evaporate, and open standards anyone could pick up. The explorers survived, so the system climbed again, higher. The internet was overvalued in 2000 and badly underestimated by 2002.
The railway manias tell the same story in iron instead of fiber. Prices collapsed and investors were wiped out, but the track, the engineering knowledge, and the operating know-how stayed in the ground and in people's heads. The technology kept advancing long after the financiers had given up on it. Electricity ran through its own waves of hype and letdown before it became the floor under everything else. Early car companies died in batches; the automobile took over the planet. In each case the bust did not destroy the higher peak. It made the market lose sight of the peak for a while, and the climb resumed as long as the people and knowledge needed for it were still around.
The sharp prediction
The crash is not the story. The aftermath is.
This reframing yields a prediction the mainstream does not make. The most severe busts will often arrive right before the most important breakthroughs.
The logic is almost mechanical. A bust lowers everyone's commitment to the existing peak. Resources come loose. Orthodox assumptions lose their grip. Incumbents pull back, and new entrants, working cheaply with newly available talent and tools, try things that were unfundable at the top of the boom. In the language of the landscape, the bust shoves the system off its local maximum and into the valley it would never have entered on purpose. The descent it spent years avoiding gets forced on it, and only from down in the valley can the climb toward higher ground begin.
That happy ending is conditional, and the condition is the whole point. A bust frees a system only if the capacity to explore survives it. If the crash destroys the explorers, the small diverse bets, the specialized talent, the shared infrastructure, and leaves the field to a few incumbents defending the old design, then the bust frees no one. It deepens the trap, now with less variety than before.
So the most useful thing this theory has to say is about what to protect. Through a bust, guard the explorers. Keep talent in the field instead of scattered to other industries. Keep infrastructure cheap and open instead of locked up by whoever is left standing. Keep a range of approaches alive even when consolidation is the easy path. The height of the next peak depends less on how violent the crash was than on how much exploratory capacity walked out of it intact.
The live case
What this means for the coming argument about AI
Plenty of careful people expect an AI bust. The usual debate asks whether AI is overhyped, whether the capabilities will disappoint, whether the valuations hold up. Those are fair questions. They are not the most revealing ones.
The local-maxima view asks something else. Is the current approach to AI nearing a local peak, and is the system preserving its ability to explore past that peak or starving it? Overvaluation and a local maximum are different diagnoses, and they call for opposite responses. One says wait for the price to correct. The other says the price may be fine while the search quietly dies.
You can lay out what each view expects. On the conventional account, AI firms are simply worth too much, the technology underdelivers, and a correction follows. On the local-maxima account, the leading approach could be nearing the top of its particular hill while money, talent, and attention pile up around it; research drifts from genuinely new directions toward scaling and tuning what already works; the spread of approaches anyone will fund narrows; real capability gains slow even as expectations, set during the steep part of the climb, keep rising. The gap opens, and a bust follows.
Here is the part worth holding onto. If an AI bust comes, it would not show that artificial intelligence failed. It would show that one approach to it reached the top of its hill. The internet survived the dot-com crash. AI could survive an AI crash the same way, as long as the explorers survive it. A crash would mark the exhaustion of a path, not of the project.
Which story are you in?
Watch the spread of approaches being funded, not just the dollar totals. Watch whether the genuine capability frontier is still moving or mostly being refined. Watch whether newcomers with different ideas can still get oxygen, or whether everything is consolidating around a few incumbents and one design. Concentrating money is what a boom looks like. Concentrating approaches is what a trap looks like.
Falsifiability
How to prove this wrong
A theory that fits every outcome explains none of them, so it is worth being clear about what would sink this one. The local-maxima view makes claims you can check against data. It predicts that in transformative sectors, the share of effort going to genuine exploration falls measurably before the bust, not after. It predicts that the real capability frontier decelerates before valuations peak, so the slowdown in substance leads the slowdown in price. And it predicts that how high a sector climbs after a bust depends on how much exploratory capacity survived, which means consolidation-heavy busts should be followed by longer stagnation than busts that leave talent and infrastructure widely available.
If busts routinely strike while exploration is still broadening and the frontier is still accelerating, the theory is wrong. If post-bust recoveries show no relationship to what survived the crash, the theory is wrong. These are not rhetorical concessions; they point at measurable things: the diversity of approaches under serious funding, the share of research budgets aimed at new directions rather than refinement, the entry rate of genuinely different newcomers, the movement of the capability frontier itself. A collapse in the variety of approaches, while the money is still flowing, is the fingerprint this theory tells you to look for. The mainstream models are not watching that gauge.
In closing
The institutionalization of success
The local-maxima theory recasts boom and bust as a problem of getting stuck, not a problem of going too far. Many crashes are not technologies overreaching their potential. They are societies, firms, and investors settling onto a real but temporary peak and mistaking it for the summit. Applied to AI, the view suggests that a bust, if it comes, need not mean the failure of the field. It can mean the exhaustion of one approach and the start of the search for a higher one.
The danger this theory points to is not ambition. It is the quiet institutionalization of success: the moment a system becomes so good at what it already does that it can no longer afford to look for what it might do next.
Comments
Post a Comment