When Small Things Become Unstoppable
How Brian Arthur’s most dangerous idea explained why markets trend, and why the profession tried to bury it
In 1983, W. Brian Arthur finished a paper that he believed would change economics.
He was not being grandiose. The paper made a modest, precise claim: in certain markets, small advantages compound. A product that gets slightly ahead attracts more users, which makes it more attractive, which attracts more users still. The process feeds on itself. Advantage begets advantage. Arthur called this increasing returns, and he demonstrated mathematically that in markets governed by this dynamic, the outcome is not determined by the quality of the competing products. It is determined by the sequence of early events: who adopted first, which product happened to gain a slight lead, what small historical accidents tipped the balance.
The implication was stark. In a world of increasing returns, the best product does not necessarily win. History matters. Small events are not averaged away. And the market does not converge on a single, predictable, optimal equilibrium. It locks in.
Arthur submitted the paper to the American Economic Review.
It was rejected.
He revised and submitted it to the Quarterly Journal of Economics.
Rejected.
He submitted it to the American Economic Review again, which had changed editors.
Rejected.
He submitted it to the Economic Journal in England.
Rejected. Then, on appeal, finally accepted. The paper was published in 1989, six years after it was written. It has since accumulated over twelve thousand citations. It is now considered one of the foundational contributions to modern economics.
But in 1983, the rejection letters were brutal. One reviewer wrote that increasing returns “cannot be an equilibrium.” Another said the paper could not be categorised as a solution to any standard, accepted economic problem. Arthur had not made an error. He had described a reality that the profession’s framework was structurally incapable of seeing.
When he later presented the ideas at Harvard, the economist Richard Zeckhauser told him bluntly: “If you are right, capitalism can’t work.” Several months later, Arthur presented the same paper at a gathering in Moscow. An equally distinguished Russian economist declared with equal vehemence: “Your argument cannot be true!”
The capitalist said increasing returns would break markets. The communist said it would break central planning. Both were correct. Increasing returns breaks any framework that assumes a single optimal outcome.
Including equilibrium.
To understand why Arthur’s idea was so threatening, you need to understand what it replaced.
Standard economics is built on diminishing returns. Dig a copper mine: the first seams are rich, the deeper ones poorer. Plant wheat: the best fields are planted first, the marginal ones later. Build a factory: beyond a certain size, coordination costs rise faster than output. In a diminishing-returns world, markets self-correct. If one firm gets too big, it runs into rising costs and competitors catch up. If one product dominates, its very dominance creates openings for alternatives. The system tends toward a predictable balance. Multiple products share the market. Prices settle. Equilibrium.
This was the economics that Alfred Marshall codified in the 1890s, and it still fills the textbooks. It was valid for the bulk-processing, smokestack economy of his day. Coal, steel, grain, rail: industries where physical constraints impose natural limits on growth.
But Arthur noticed that a different kind of market had emerged, one where the constraints ran in reverse. In knowledge-based industries, in network technologies, in platform economies, the costs were overwhelmingly upfront: research, design, development. Once the product existed, the marginal cost of each additional unit was close to zero. And crucially, each additional user made the product more valuable to every other user.
This is increasing returns. The more users adopt a product, the more reasons others have to adopt it. The more software is written for an operating system, the more people buy computers that run that system. The more people join a social network, the more valuable the network becomes. The more traders use a particular exchange, the tighter the spreads, the more traders use it.
In a diminishing-returns world, the outcome is determined by the fundamentals: who has the best product at the lowest cost. In an increasing-returns world, the outcome is determined by the dynamics: who got ahead first, and how fast the feedback loop accelerated.
Fundamentals versus dynamics. Prediction versus feedback.
That distinction will matter enormously when we get to markets.
The technology industry of the late twentieth century was Arthur’s laboratory. And it proved him right in the most spectacular fashion.
Consider the videotape format war of the early 1980s. Two systems competed: Sony’s Betamax and JVC’s VHS. By most technical measures, Betamax was superior: better resolution, more compact cassettes, cleaner image quality. Sony launched first, in 1975. JVC followed with VHS in 1976.
Under diminishing returns, the superior product should have won. Under increasing returns, what mattered was the feedback loop. JVC licensed its technology to multiple manufacturers, which meant more VHS players on shelves, which meant video rental stores stocked more VHS titles, which meant consumers bought more VHS players, which meant studios released more titles on VHS. The loop accelerated. By 1988, VHS held over 90% of the market. Sony surrendered and began manufacturing VHS machines itself.
Betamax was better. VHS won. Not because the market was irrational, but because the market was governed by a different kind of logic: the logic of positive feedback, where small early advantages compound into lock-in.
The operating system wars told the same story at a larger scale. In the early 1980s, three systems competed for the personal computer market: CP/M (the early leader, established by 1979), Apple’s Macintosh (elegantly designed, easy to use), and DOS (Microsoft’s offering, born from a deal to supply IBM’s personal computer). Operating systems exhibit increasing returns in their purest form: if one system gets ahead, developers write more software for it, which attracts more users, which attracts more developers.
DOS was not the best system. It was derided by professionals. But it shipped on every IBM PC, and IBM’s corporate credibility gave it an early installed base. The feedback loop did the rest. By the early 1990s, Windows (the graphical successor to DOS) had locked in the market. Microsoft could spread its development costs over hundreds of millions of users. The margins were extraordinary. Not because of superior quality, but because of the dynamics of increasing returns.
Eric Schmidt, then chief technology officer at Sun Microsystems, would later say: “We launched Java based on Arthur’s ideas.” The entire strategy of the 1990s technology economy, from Netscape to Amazon to Google, was built on the recognition that in network markets, getting ahead early and accelerating the feedback loop mattered more than having the best product.
Arthur’s paper had been rejected because increasing returns “cannot be an equilibrium.” The technology industry of the next three decades proved that the economy didn’t care.
Now translate this into financial markets.
In The Fractals of Finance, I described the self-similar geometry of price: the way that patterns at one timescale echo at another, the way that volatility clusters, the way that markets exhibit memory. These are not random phenomena. They are the fingerprints of positive feedback in price.
When an asset’s price begins to rise, the rise itself creates the conditions for further rise. Rising prices attract momentum capital. Momentum capital pushes prices higher. Higher prices generate media coverage, analyst upgrades, and investor enthusiasm. Enthusiasm attracts more capital. The loop feeds on itself in exactly the same way that VHS adoption fed on itself, or that Windows installation fed on itself. The dynamics are identical. The medium is different.
We do not call this increasing returns. We call it a trend.
A trend is increasing returns operating in the domain of price. It is positive feedback made visible on a chart. And like every increasing-returns dynamic Arthur described, it exhibits the same three properties: path dependence (the current price depends on the sequence of events that produced it, not on some abstract “fair value”), lock-in (once a trend establishes, it persists longer than fundamentals alone would justify), and non-ergodicity (the outcome depends on which path was taken, and you cannot rerun the tape of history and expect the same result).
Under equilibrium theory, a price that has risen “too far” above its fundamental value must revert. The market is a negative-feedback system: deviations from equilibrium trigger corrections. Under increasing returns, a price that has risen continues to attract the forces that make it rise further. The market, in this regime, is a positive-feedback system: deviations from equilibrium are amplified, not corrected.
Both dynamics exist. This is the crucial point. Markets are not purely one or the other. They alternate between regimes of positive feedback (trends) and regimes of negative feedback (reversions), and the transitions between them are neither smooth nor predictable. In Complex Adaptive Markets (forthcoming 2026), I describe this as the breathing of the market: expansion and contraction, trend and reversion, increasing returns and diminishing returns, each giving way to the other in patterns that are recognisable in aggregate but unpredictable in their specifics.
Here is where Arthur’s increasing returns connects directly to the framework we established in the previous article.
Recall the distinction between predictive and responsive strategies. Predictive strategies forecast a specific outcome: this stock is overvalued, this currency will weaken, this technology will win. Responsive strategies observe what the system is doing and align with it: the price is trending, the spread is stretching, the regime is shifting.
Increasing returns is the mechanism that makes predictive strategies so dangerous and responsive strategies so durable.
Consider the videotape format war from a trader’s perspective. A predictive strategy in 1980 would have said: “Betamax is technically superior. It will win the format war. Buy Sony.” This prediction was based on a careful analysis of fundamentals: resolution, size, image quality. It was the kind of analysis that a diminishing-returns world rewards. And it was wrong, because the market was governed not by fundamentals but by feedback.
A responsive strategy would not have predicted which format would win. It would have observed the dynamics. When VHS began to pull ahead in adoption, the responsive strategy would have noted the acceleration and aligned with it. It would not have needed to know why VHS was winning. It would only have needed to see that it was winning, and that the feedback loop was strengthening.
This is not a trivial distinction. It is the difference between a strategy that bets on what should happen (based on a model of value) and a strategy that aligns with what is happening (based on the observable dynamics of feedback). In an increasing-returns world, the first strategy is systematically wrong whenever the feedback loop overpowers the fundamentals. The second strategy is systematically aligned with the most powerful force in the market.
Arthur’s paper was rejected because increasing returns “cannot be an equilibrium.” The reviewers were correct: it cannot. But their error was in assuming that equilibrium was the relevant standard. In a market where positive feedback drives prices, equilibrium is not a useful concept. Feedback is. And a strategy that follows feedback does not need equilibrium to justify itself.
But there is a second half to the story, and it is equally important.
Lock-in is not permanent. It only feels permanent while it lasts.
VHS dominated for two decades, then was swept away by DVD. Windows locked in the desktop, then mobile computing arrived and the lock-in became irrelevant. Kodak dominated film photography for a century, and the digital transition destroyed it in a decade. Every lock-in is eventually broken, not by a competitor within the same feedback loop, but by an external shift that makes the entire loop obsolete.
In financial markets, this moment, when a locked-in regime breaks, is among the most violent events a trader can experience.
On January 15, 2015, the Swiss National Bank abandoned its three-year policy of capping the Swiss franc at 1.20 against the euro. For three years, the cap had been the dominant fact of the EUR/CHF market. An entire ecology of predictions had formed around it: carry traders borrowing in francs and lending in euros, corporate treasurers hedging on the assumption that the floor would hold, options markets pricing in the SNB’s commitment. The prediction (“the franc will not strengthen beyond 1.20”) had achieved the status of certainty. It was, in Arthur’s terms, locked in.
When the SNB removed the cap without warning, the franc appreciated by nearly 30% against the euro in minutes. Not hours. Minutes. Broker after broker reported negative client equity. Several firms went bankrupt. The locked-in prediction (“the floor will hold”) did not merely fail. It generated a move of historic violence in the opposite direction, precisely because the lock-in had been so complete that virtually no one was positioned for the alternative.
This is increasing returns in reverse. The unwinding of a locked-in position generates a feedback loop of its own: selling begets selling, margin calls beget margin calls, the exit becomes as self-reinforcing as the entry. The trend that follows a lock-in break is not an anomaly. It is the structural mirror image of the trend that built the lock-in in the first place.
I wrote about this in The Fractals of Finance: boundaries store energy. The longer the compression, the more violent the release. The Swiss franc de-peg was a textbook case. Three years of stored energy, released in minutes. The predictive strategies that had assumed the lock-in would hold were annihilated. The responsive strategies that observed the directional move and followed it captured one of the largest currency moves in modern history.
There is a lock-in forming right now that Arthur’s framework illuminates with unsettling clarity.
The rise of passive investing is the purest example of increasing returns in contemporary finance. When BlackRock’s iShares launched the first wave of exchange-traded funds, the feedback loop was immediate: the largest funds attracted the most volume, which produced the tightest spreads, which attracted more investors, which made the funds larger. By 2024, the top three S&P 500 ETFs held combined assets exceeding $1.5 trillion. Index funds as a whole controlled more than half of all US equity fund assets.
Under diminishing returns, this dominance would self-correct. Competitors would emerge. Costs would equalise. Market share would distribute. Under increasing returns, the dominance accelerates. Every new dollar of passive investment flows disproportionately to the largest funds, which hold the largest stocks, which pushes those stocks higher, which increases the index weight of those stocks, which means the next dollar of passive investment allocates even more to them.
This is not a prediction about where the stock market is going. It is an observation about the dynamics that are operating right now. The feedback loop between passive flows, index weights, and stock prices is increasing returns in real time. And it has consequences that no equilibrium model is equipped to describe.
Because every lock-in eventually breaks. The question is never whether the feedback loop will reverse. It is when, and what will happen to the ecology of predictions that formed around the assumption that it would not.
When (not if) the passive flow regime shifts, the unwinding will be governed by the same increasing-returns dynamics that built it, operating in reverse. Selling will beget selling. Stocks that rose because of mechanical inflows will fall because of mechanical outflows. The move will be larger than any fundamentals-based model would predict, because the lock-in was driven by feedback, not fundamentals.
A predictive strategy that says “passive is the new normal and will continue indefinitely” is making exactly the kind of bet that increasing returns punishes. It is predicting that the current lock-in will hold. Every lock-in holds until it doesn’t.
A responsive strategy will not predict the timing of the break. It will observe the break when it occurs and align with the feedback, whatever direction it takes.
Arthur’s increasing returns gives us the deepest explanation I know for why trend following works and why its edge does not decay.
Trends exist because positive feedback exists. Positive feedback exists because markets are populated by agents who react to each other’s behaviour, creating the self-reinforcing dynamics that Arthur spent his career studying. This is not a market inefficiency. It is the fundamental operating logic of a complex adaptive system. You cannot arbitrage it away any more than you can arbitrage away gravity.
And the responsive strategy’s advantage is structural, not informational. The trend follower does not need to know which product will lock in the market. The trend follower does not need to predict which regime will break. The trend follower observes the feedback and aligns with it: riding the increasing-returns phase when it builds, cutting when it reverses, and positioning across enough markets (as Niels Kaastrup-Larsen and I detail in Trend Following Manifesto, forthcoming 2026) that wherever the next feedback loop is forming, the programme is there to capture it.
Mean reversion occupies the structural complement. When an increasing-returns regime overshoots (and they always overshoot, because positive feedback has no internal brake), the snap-back is the domain of the mean reversion strategy. Not because mean reversion “predicts” a specific fair value. But because it observes the extreme and aligns with the restoring force when the feedback exhausts itself.
Both are structural responses to the dynamics that increasing returns produces. Neither is a prediction that can be crowded out. Neither requires equilibrium to justify itself. Both persist for the same reason that trends and reversions persist: because positive feedback is a permanent feature of markets populated by adaptive, reactive, self-reinforcing agents.
In Carved by Impossibility (forthcoming 2026), I argue that robust structure is what remains after everything fragile has been eliminated. Increasing returns eliminates the fragile prediction that “the best product wins.” It eliminates the fragile assumption that “prices reflect fundamentals.” What remains, the durable structure, is feedback itself. And the strategies that align with feedback rather than fighting it are the ones that survive.
Arthur once wrote that in an increasing-returns world, historical small events are not averaged away and forgotten by the dynamics. They may decide the outcome.
Think about what that means for trading.
Under equilibrium theory, the past is irrelevant. Prices reflect current information. History is already priced in. Under increasing returns, the past is everything. The sequence of events matters. The path matters. A small event, a rumour, a policy announcement, a single large order, can trigger a feedback loop that compounds into a move that no fundamental model predicted and no equilibrium framework can explain.
This is not randomness. It is path dependence. The outcome is determined by the path the system took, and the system could have taken a different path and arrived at a different outcome. The QWERTY keyboard could have been replaced. VHS could have lost. Windows could have fallen to the Macintosh. The Swiss franc floor could have held for another decade. History is not destiny. But once the feedback loop locks in, it behaves as though it is.
The responsive trader understands this intuitively. You do not fight the feedback. You do not argue with the tape. You follow the path the market has chosen, because in an increasing-returns world, the path is the signal.
Arthur waited six years for the economics profession to accept this. The market has been demonstrating it every day for a century.
Follow the feedback.
Next in the series: “All Systems Will Be Gamed.” Arthur’s theory of exploitation and emergent failure, the 2008 crisis as a complexity event, and why the programme that assumes boundaries will break is the one that survives.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets
The book explores the full architecture of feedback, fat tails, and fractal structure in financial markets, and what it means for how we trade, invest, and understand risk.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.