How Smart Money Blows Up

The arithmetic that ends a sophisticated book

Leverage does one thing: it divides the distance between a normal move and a fatal one. A five-times-levered book is wiped out by a 20% fall and halved by a 10% one; a twenty-five-times-levered book is wiped out by 4% and halved by 2%. Neither number requires a forecast of catastrophe. A 4% decline in a diversified equity portfolio is a quiet Tuesday, and a 20% decline is an ordinary correction. What turns those into terminal events is the borrowing: the debt is fixed while the assets move, so the entire move lands on the equity, which is the same mechanism as the maintenance call in R8 scaled up. Concentration is the second ingredient, and it is what makes the fall happen to the whole book at once. A broadly diversified five-times-levered portfolio can survive a great deal, because only part of it is exposed to any one shock. A book concentrated in a handful of related positions has one loss, not several: when the theme moves, every position moves with it, the equity is hit by the sum, and the margin call arrives with no diversification to draw on. That is the R10 cluster problem with a multiplier attached, and it is the reason both celebrated failures were concentrated rather than diversified. The third ingredient is the mechanism of failure itself, which is not the loss but the **forced sale**. Margin agreements require collateral daily, and when the collateral is the position, meeting a call means selling into the same decline that generated it. Other holders see the forced flow, the price falls further, the collateral requirement rises, and the process feeds itself. This is why the outcome exceeds what the initial move would suggest and why it can happen to a book whose underlying thesis is partly correct: the value of LTCM portfolios, and several of Archegos positions, were not obviously wrong. Nothing in the failure depended on the analysis being bad; it depended on the structure being unpayable. The fall that ends the book — Five-times leverage: wiped out by a 20% fall, halved by 10% · Twenty-five-times leverage: wiped out by a 4% fall, halved by 2% ← · What concentration changes: the whole book takes the same loss at once · What margin does: converts the loss into forced selling that deepens it The famous failures are usually described as intelligence defeated by markets. The mechanism is less flattering and easier to guard against: leverage plus concentration plus daily collateral turns a survivable move into an unsolvable one.

Four lessons that generalise

The first is that the **size of the fall that matters is set by the structure, not by the thesis**. If a book is wiped out by 4%, then the risk analysis has to be about whether 4% can happen, which it obviously can, rather than about whether the thesis is right over three years. Ask what move finishes the position before asking whether the position is a good idea. The second is that **being right is not a plan**. Both failures contained positions that paid off eventually or were substantially fair-valued; the holders did not get to wait. Any strategy that requires the market to agree within a deadline is a strategy with a funding risk, and the funding risk is usually invisible in the backtest because backtests have no margin calls. The third is that **correlation in a crisis is not the correlation you estimated** — the R10 argument — and it applies with particular force to a levered book, because every position is financed by the same collateral pool. Two uncorrelated positions financed together are not two bets when a call arrives; they are one decision about how much to sell. The fourth is the household version of the same idea, which is smaller and just as real. Leverage does not have to be margin debt to appear: a mortgage against an investment portfolio, a car loan funded while investing, an emergency fund spent to buy more of a falling position. Each is a funding arrangement with a call date, and each converts a temporary fall into a permanent one. The lesson is the same arithmetic at a smaller scale. The household version — Borrowing against a portfolio: a maintenance call on your own balance sheet · Investing the emergency fund: removing the buffer that funds a bad year · A loan payment funded by a volatile asset: a call date you did not choose ← · The general rule: no position should require someone else permission to survive The characteristic mistake of a sophisticated investor is not a bad estimate; it is a structure that cannot wait for a good one. The check is short — what move would force a sale, and when — and it takes five minutes.

The same failure, five times

The academic and journalistic literature on large losses converges on a short list of episodes, and the striking thing about reading them together is how much of the anatomy is identical. A fund earns a reputation for sophistication, runs a strategy that is genuinely clever, and then adds the three ingredients that convert clever into fragile: **leverage**, **concentration**, and a **funding structure that someone else can withdraw**. The strategy usually works for years; the failure is rarely a wrong view but a correct view held with more size than the funding could survive. Consider the shape rather than the names. A relative-value fund in the late nineties was levered roughly twenty-five to one on the theory that spreads would converge; the positions were sound and the funding was short-term, so when the market moved against the spread and the lenders asked for more collateral, it had to sell into the move — which is the definition of the failure, and it required a consortium rather than a liquidation to stop it. A natural-gas trader in the mid-2000s held an enormous calendar spread in a market with limited depth; the position was eventually right and the margin call arrived first. An options-writing exchange-traded product in 2018 held a short-volatility position that had been profitable for years and needed one session to end, because the product’s own rebalancing fed the move that destroyed it. A family office in 2021 ran concentrated long positions funded by prime-broker margin and swap-based leverage that did not appear on anyone’s screens until the collateral ran out. And the 2022 credit unwind took several crypto lenders and a large fund with the same structure at once. Five episodes, and each one adds a specific mechanism the others lack: **crowded trades** that make the exit itself the loss, **positions that are right at a horizon longer than the funding**, **products whose rules mechanically force them to trade in the direction of the move**, **leverage that is invisible because it is synthetic**, and **collateral that is accepted at a valuation that only holds while everyone agrees**. That is the useful map, and it generalises well enough to be checked against any book: where is the size relative to the market’s depth, how long can the position be held if it goes against me, would my own rules force me to sell at the bottom, is the leverage reported, and what is my collateral worth if the market stops agreeing with the price. • Leverage, concentration and withdrawable funding, in the same order, every time. • Crowded trades turn the exit into the loss even when the view is right. • Mechanical rules (rebalancing, margin, redemption) can force selling into weakness. • Synthetic leverage and optimistic collateral values are invisible until the call arrives.

How to tell when you are the crowd

The cases in this lesson share a shape: a position that looked diversified, well-hedged and modest in size, held by many people who had all reached it by the same reasoning. The single most transferable question to ask of your own book is whether it is *crowded* — and crowding is observable, in a handful of indirect ways, before it becomes a stampede. The clearest signal is the price of the exit. If a position can only be built or reduced by paying a wide spread, and the spread widens further whenever the idea becomes popular, the market is telling you the trade is one-sided. In options, a persistent skew or an elevated implied volatility on the side you are selling is the same message from a different instrument. So is an unusually high borrow fee for a short: it means more people want to borrow the shares than there are shares to lend, which is a position count rather than an opinion. The second signal is the company your P&L keeps. Track how much of your book’s daily variance is explained by a single factor — one theme, one rate, one commodity, one liquidity condition. A collection of names that all depend on rates falling is a single trade, however many different tickers it uses, and it will not feel like one until it is unwound. This is the same point the correlation section of this subject makes, applied to the question of crowding rather than to the question of diversification. The third is the quality of the argument. Crowded trades are usually supported by a case that has become easy to state — a compelling narrative, a widely published valuation, a consensus view repeated in the same words by people who have not independently done the work. The tell is not that the argument is wrong. It is that the argument is arriving from everywhere at once and is no longer being tested, which is what makes the marginal buyer scarce. And the fourth is the asymmetry of the exit. A crowded position is one where everyone’s stop is at a similar place, because everyone watched the same level. That is what turns an ordinary decline into a cascade: the first break triggers the crowd’s stops, the selling pushes the price further, which triggers the next tier. The defence is not to avoid popular ideas — the well-known valuations are often the correct ones — but to know what the exit looks like if the crowd leaves at once, and to size the position so that being early by a quarter can be survived. The practical test is a sentence: if this works and everyone has it, what happens when a few of them get out? If the honest answer is a disorderly exit that you would be part of, the position is a crowding risk, and the size is the only lever that answers it. • A widening spread or a rising borrow fee is the market pricing a one-sided trade. • Measure how much of your variance comes from one factor, not from how many tickers you hold. • A compelling case repeated in the same words everywhere has stopped being tested. • Crowded positions share the same stop level, which is what makes the exit disorderly. The most useful habit the case studies offer is not a checklist but a question asked before the position is opened: who is on the other side of this, and what happens to my exit if they all decide at once? A trade with no answer to that has an unexamined risk in it.

Sources

• The collapse of Long-Term Capital ManagementPresident’s Working Group on Financial Markets, "Hedge Funds, Leverage and the Lessons of Long-Term Capital Management" (1999)
• The Archegos default and its prime-broker exposuresCredit Suisse and Nomura disclosures; regulatory reports on the March 2021 Archegos default
• Funding liquidity and forced selling spiralBrunnermeier & Pedersen, "Market Liquidity and Funding Liquidity" (2009)
• Crowded trades and the cost of being right with everyone elseStandard literature on crowded positioning and unwind risk