The Bar Problem That Explains Every Market
Why predictive strategies destroy themselves in reflexive markets, and why the edge that feeds on them never decays
There is a bar on Canyon Road in Santa Fe, New Mexico, called El Farol. It is a small place: thick adobe walls, dim lighting, the kind of room where a crowd of sixty feels like a party and a crowd of a hundred feels like a punishment. On Thursday nights in the early 1990s, they played Irish music there. People would go if they expected it to be pleasant. They would stay home if they expected it to be packed.
W. Brian Arthur, the Irish-born economist who had helped launch complexity economics at the Santa Fe Institute, lived a few blocks away. He walked past El Farol regularly. And one evening, watching the unpredictable ebb and flow of the Thursday night crowd, he recognised something that had been nagging at him for years.
The bar had a problem that economics could not solve.
Here is the setup. One hundred people want to go to El Farol on Thursday night. But the bar is only enjoyable if fewer than sixty show up. If more than sixty come, everyone has a miserable time and wishes they had stayed home. Each person must decide, independently and simultaneously, whether to go. They cannot call ahead. They cannot check a live webcam. They have only one input: the history of past attendance.
The question Arthur posed was deceptively simple: what strategy should each person use to decide?
The answer, it turns out, is that there is no correct strategy. And that single insight, published in the American Economic Review in 1994, would quietly reshape how we think about competition, crowding, and survival in financial markets.
The reason there is no correct strategy is not because the problem is hard. It is because the problem is reflexive.
Suppose someone discovers that a particular forecasting rule works beautifully: “go if last week’s attendance was below forty.” If that rule works, word spreads. More people adopt it. But now, every week that attendance falls below forty, everyone using the rule goes to the bar. The bar becomes packed. The rule stops working precisely because it worked.
This is not a flaw in the specific rule. It is a structural feature of any system in which predictions alter the reality being predicted. If every person in Santa Fe adopted the same strategy (any strategy, however sophisticated), it would be guaranteed to fail. A strategy that predicts low attendance causes high attendance. A strategy that predicts high attendance causes low attendance. The act of believing the forecast destroys the forecast.
Note the key word: predicts. Every strategy in the El Farol problem is a prediction about a specific future state. “The bar will be empty.” “The bar will be full.” “Attendance will mirror last week’s pattern.” Each agent is committing to a forecast and acting on it. It is the predictive nature of these strategies that makes them vulnerable. When too many agents share the same prediction, the prediction becomes its own refutation.
This is reflexivity in its purest form. George Soros popularised the term in finance, but Arthur built the mathematical demonstration. In a reflexive system, the participants’ beliefs do not merely reflect reality. They shape it. And because they shape it, any belief that becomes sufficiently widespread will reshape reality into something that contradicts the belief.
Arthur called this “inductive reasoning under bounded rationality.” But what he was really describing was something more fundamental: a world in which prediction, by its very nature, is self-defeating at scale. No equilibrium. No resting point. The bar’s attendance oscillates perpetually, never settling, always adapting, forever in motion.
In his simulation, Arthur gave each of the hundred agents a small collection of forecasting strategies drawn from a larger pool: some extrapolated from last week, some averaged the past four weeks, some used contrarian logic, some looked for cycles. Each agent used whichever of their strategies had been most accurate recently. When a strategy failed, they rotated to another. No agent could see what other agents were doing. No agent had a “god’s-eye” view.
What emerged was an ecology.
This is the word Arthur used, and it is the right one. Not a competition. Not a game with a winner. An ecology: a living, shifting population of predictive strategies that co-evolve, feed on each other, and keep the system in a state of perpetual disequilibrium.
Momentum predictions would do well for a stretch. Attendance would rise in streaks, and agents using trend-extrapolation rules would correctly forecast the bar’s popularity. But as more agents rotated into momentum predictions, the bar would overshoot. Then contrarian predictions would outperform: agents who predicted the opposite of the crowd would stay home during packed weeks and visit during empty ones. Contrarians would thrive for a while, until enough agents adopted the same contrarian prediction that it became the consensus, at which point it too would fail.
The system never rested. No single predictive strategy dominated permanently. The ecology was self-correcting: any prediction that attracted too many followers generated the conditions for its own failure, which opened space for alternative predictions, which in turn attracted followers, which started the cycle again.
When I first encountered the El Farol problem, years before I wrote The Fractals of Finance, I remember the sensation of recognition. Not the intellectual kind, where you nod at an elegant proof. The physical kind. The gut-level feeling of: this is the market. This is what I have been trading all along.
But it took me much longer to see the deeper implication. The El Farol problem does not merely describe a world where strategies fail. It describes a world that sorts strategies into two fundamentally different categories. And everything in trading follows from understanding which category you are in.
Every financial market is El Farol. Every active participant is carrying a rule that maps available information onto a decision. The collective action of all these rules produces the price. And the price, in turn, is the very information that the rules are trying to forecast. This is the reflexive loop.
But here is the critical distinction, and it is one I have come to believe is among the most important in all of trading.
Some strategies are predictive. They forecast a specific state of the world and bet on it. “GameStop will go to zero.” “Volatility will remain low.” “The euro will hold this range.” “This stock is worth $45 based on my discounted cash flow model.” These are the strategies that populate the El Farol ecology. They are making a claim about what the future should look like. They commit capital to that claim. And they are subject to the fundamental El Farol dynamic: when enough capital crowds into the same prediction, the prediction undermines itself.
Other strategies are responsive. They do not predict what the world will look like. They observe what the world is doing and align with it. A trend follower does not forecast that oil will reach $100. A trend follower observes that oil is rising, enters the position, and exits when the rise stops. A mean reversion strategy does not predict that a spread will close. It observes that a spread has stretched beyond a historical norm and positions for a return to that norm. Neither strategy requires a view on what should happen. Both require only that the market is exhibiting a recognisable dynamic.
This distinction is not semantic. It is structural. And it determines everything about how a strategy relates to the El Farol ecology.
Predictive strategies are the ecology. They are the bar patrons: guessing, forecasting, committing to a specific view, subject to crowding, subject to reflexive self-destruction. They live inside the loop.
Responsive strategies sit outside the ecology. They do not participate in the prediction game. They watch the ecology from the window. And when predictive strategies crowd, overshoot, and collapse (as they inevitably must in a reflexive system), responsive strategies harvest the directional moves and reversions that the collapse produces.
Predictive strategies are the food. Responsive strategies do the feeding.
On the morning of January 25, 2021, shares of GameStop, a struggling American video game retailer, opened at $96.73. Two weeks earlier they had been trading at $20. Two months earlier, at $12. By January 28, the stock would touch $483 in the pre-market.
The surface narrative was simple: a group of retail traders on the Reddit forum WallStreetBets had identified that several large hedge funds held enormous short positions in GameStop. The retail traders began buying shares and call options en masse, forcing the short sellers to cover their positions by buying more shares, which pushed the price higher, which forced more covering, which pushed the price higher still.
The deeper narrative was El Farol. And the distinction between predictive and responsive strategies was on full display.
Consider the hedge funds. Their strategy was predictive: GameStop is a declining business in a digital world. It is worth less than the current price. Short it, collect the premium as the share price drifts toward zero. This was a reasonable prediction when only a few funds held it. But it was not a few funds. By late 2020, the short interest in GameStop exceeded 140% of the available float. More shares had been sold short than actually existed.
One hundred and forty per cent. Think about that number in El Farol terms. It is not that the bar was merely full. More people had promised to leave the bar than were actually in the bar. The prediction (“this stock is going to zero”) had attracted so much capital that it had created a structural impossibility. The crowd of predictors was so large that the exit could not physically accommodate them all.
The Reddit traders did not invent the fragility. They merely identified it. They saw, with the intuitive clarity of the outsider, that the ecology of predictions had become dangerously homogeneous: too many participants making the same forecast, relying on the same thesis, exposed to the same risk. When the retail buyers applied pressure, the ecology of predictions collapsed. The dominant forecast (“GameStop goes to zero”) produced exactly the outcome it had not modelled: a violent, self-reinforcing squeeze in the opposite direction.
And that squeeze? That violent directional move? That was a trend. It was the structural consequence of prediction failure in a reflexive market. It was food.
A responsive strategy did not need to know anything about GameStop’s business model, its short interest, or the Reddit forums. It needed only to observe that the price was moving with persistence and follow it.
If GameStop was the El Farol ecology collapsing over weeks, Volmageddon was the same dynamic compressed into hours.
By early 2018, one of the most popular predictions in global markets was that volatility would stay low. The VIX had spent most of 2017 below 12, near historic lows. Products designed to profit from this prediction (inverse volatility ETPs like the VelocityShares Daily Inverse VIX Short-Term note, ticker XIV) had delivered extraordinary returns. XIV alone held $1.9 billion in assets. The prediction felt like fact. Everybody believed it.
Everybody was going to the bar.
On February 5, 2018, the S&P 500 dropped 4.1%, its largest single-day fall since 2011. In isolation, a painful but unremarkable event: such drops happen roughly twice a year on average. But the ecosystem of predictions built on top of the volatility complex was not designed for unremarkable. It was designed for calm.
The VIX surged from 17 at the open to 37 at the close, a gain of more than 100% in a single session: the largest daily point increase ever recorded. XIV’s value dropped from $1.9 billion to $63 million between the opening and closing bells. Ninety-seven per cent of its value, gone.
But the real violence was structural. The inverse volatility ETPs were required to rebalance their hedges at the close of each trading day. As volatility spiked, they needed to buy VIX futures to maintain their target exposure. Buying VIX futures pushed the VIX higher. A higher VIX meant more losses on the inverse products, which meant more rebalancing, which meant more buying, which meant a higher VIX. The reflexive loop turned mechanical.
The ecology of predictions had collapsed. The dominant prediction (“volatility will stay low”) had attracted so much capital that the rebalancing mechanics of the products themselves became the source of the very volatility they were betting against. The predictors caused the outcome they had predicted could not happen.
Credit Suisse terminated XIV on February 21. The product that had been a licence to print money for years was dead in a fortnight.
And once again: the collapse of the predictive ecology generated a massive, persistent, directional move. The VIX did not just spike and revert. It shifted regime. It trended. The wreckage of a billion dollars’ worth of failed predictions became food for any responsive strategy positioned to capture it.
This brings us to the question that matters most.
Every serious trader understands alpha decay. You find an edge. Others discover it. Capital flows in. The edge gets crowded and disappears. This is the El Farol dynamic, and it applies mercilessly to every predictive strategy ever invented. Value investing worked spectacularly until everyone became a value investor. Statistical arbitrage worked until every quant fund was running the same pairs. Merger arbitrage, carry trades, volatility selling: each one a prediction about a specific state of the world, each one subject to the reflexive cycle of adoption, crowding, and collapse.
So why doesn’t trend following decay?
The question assumes that trend following is a predictive strategy whose alpha can be competed away through imitation and adaptation. It is not. Trend following is a responsive strategy whose edge is not prediction but alignment with a structural feature of reflexive markets.
The edge is not in the forecast. There is no forecast. The edge is in the response to a permanent feature of reflexive systems: that predictions crowd, collapse, and generate directional moves. This will happen for as long as there are participants in markets making forecasts about the future. Which is to say: always.
Think about it in ecological terms. A lion’s edge does not decay because the antelope learn to run faster. The relationship between predator and prey is structural. As long as there are antelope, there will be lions. As long as there are predictive strategies crowding and failing in reflexive markets, there will be trends and reversions. And as long as there are trends and reversions, responsive strategies will harvest them.
This is not decaying alpha from prediction and adaptation. This is a structural symptom of a reflexive market.
I wrote about this from a different angle in The Fractals of Finance: the way that apparent stability can mask the accumulation of hidden fragility. When the ecology of predictions becomes homogeneous, when everyone is positioned on the same side, the system does not gradually adjust. It snaps. The boundary holds, holds, holds, and then it does not. The snap is a trend. The reversion from the snap is a mean reversion. Both are structural consequences of the same underlying dynamic: prediction failure in a reflexive system.
Trend following and mean reversion are not competing strategies. They are two faces of the same structural response to reflexivity. One captures the directional energy released when predictions fail. The other captures the restoring force when the release overshoots. Both persist for the same reason: their cause is not a market inefficiency that can be arbitraged away. It is the market itself. It is what reflexivity produces, permanently, as a condition of its own existence.
But even a structural edge requires the right architecture to harvest it. And this is where the El Farol insight connects to something I am writing about extensively in Carved by Impossibility (forthcoming 2026): the idea that robust structure is not designed. It is what remains after everything fragile has been eliminated.
The El Farol ecology tells us that prediction failure is perpetual but unpredictable. We know that predictive strategies will crowd and collapse somewhere, in some market, at some time. We cannot know where, or when, or which prediction will break. The cycle is certain. Its specific expression is not.
This is why diversification is not a risk management tool. It is a structural necessity.
Think about it in El Farol terms. If predictive ecologies are perpetually forming and collapsing across markets, then the worst thing you can do is stand outside one bar and wait. The best thing you can do is spread your attention across many different bars, in many different towns, on many different nights. Some will be packed (prediction crowding). Some will be emptying out (prediction failure). Some will be in the quiet lull between cycles. The responsive strategy does not need every bar to produce a signal. It needs some bars, somewhere, to be in the process of ecological disruption.
Niels Kaastrup-Larsen and I explore this at length in Trend Following Manifesto (forthcoming 2026): the portfolio is not a collection of bets on individual markets. It is an array of responsive sensors deployed across dozens of uncorrelated price series, each of which contains its own ecology of predictions, its own crowding dynamics, its own reflexive loop. The programme does not need any individual market to trend. It needs some markets, somewhere, to be experiencing the structural consequences of prediction failure.
Over the decades, markets have always provided this. Energy markets crowd one way and then unwind. Bond markets compress and then break. Currency regimes accumulate predictive consensus and then shatter. Agricultural markets respond to supply shocks that no prediction anticipated. The diversity of these ecologies is not incidental. It is the architecture that allows a responsive strategy to remain permanently aligned with a permanent feature of reflexive markets.
When traders ask me “why so many markets?” I want to hand them Arthur’s 1994 paper and say: because prediction failure is perpetual but unpredictable. Because you cannot know which bar will overflow on any given Thursday. Because the only durable architecture for harvesting a structural edge is breadth.
There is a pattern that Arthur identified in the El Farol simulation that I find remarkable. Over time, the system self-organises so that attendance hovers around the comfort threshold of sixty people. Not because the agents coordinate. Not because anyone solves the problem. But because the ecology of predictions, through constant failure and renewal, converges on an emergent regularity that looks, from the outside, almost like intelligence.
Markets do this too. Over long periods, diversified responsive strategies produce returns that appear to come from nowhere: not correlated with equities, not correlated with bonds, not correlated with each other in any stable way. These returns are not alpha in the traditional sense. They are not the product of superior prediction. They are generated by the ecology itself: by the ceaseless process of prediction formation, crowding, failure, and renewal that is the heartbeat of every reflexive market.
This is not anomaly. It is structure.
The El Farol problem tells us that there will never be a stable equilibrium in financial markets. Predictions will always emerge, attract capital, become crowded, and fail. New predictions will replace them. The cycle is perpetual. And every turn of the cycle generates the directional moves and reversions that responsive strategies are built to capture.
The conventional language of finance calls this alpha, and worries about its decay. But alpha decays because it rests on prediction, and prediction in a reflexive system is inherently temporary. What trend following and mean reversion capture is something entirely different: a structural symptom of reflexivity itself. It does not decay because reflexivity does not decay. It is the permanent condition of a market populated by agents who are trying, always trying, and always failing, to predict each other.
As Arthur wrote: the economy is not mechanistic, static, timeless, and perfect. It is organic, always creating itself, alive and full of messy vitality.
The El Farol bar is still there on Canyon Road. The music still plays on Thursday nights. The crowd still rises and falls in patterns that no one can predict and everyone can feel.
The predictions are the crowd. The response is the signal.
Follow it.
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.