The Masterpieces Gallery
The anatomy of asymmetry
The Tulip Wing shows what crowds do to prices. This gallery shows the opposite masterpiece: trades that made history not by predicting better, but by finding structural asymmetry — bounded loss, unbounded gain, and a counterparty who had to act. Study the anatomy, not the names. None of this is advice; every one of these setups is famous precisely because it was rare.
Soros & Druckenmiller vs the pound
The UK had pegged sterling inside Europe’s exchange-rate mechanism at a level its economy couldn’t support. The peg capped how far the pound could rise; devaluation was unbounded. Reported profit when the peg broke: over a billion dollars.
Anatomy: capped downside (the band), open upside (devaluation), and a forced counterparty — the Bank of England had to defend the peg with finite reserves against the whole market. Not a prediction. A payoff shape.
Burry, Paulson & the CDS trade
Credit default swaps on subprime mortgage bonds cost a small running premium and paid face value on collapse. The bet could be wrong for years at a known, budgeted cost — and right exactly once.
Anatomy: premium-bounded loss, ~10-100x payoff, and a structural mispricing (models assumed house prices don’t fall nationally). The hard part wasn’t the idea — it was carrying the negative carry until being right. Asymmetry pays the patient and breaks the leveraged.
The Volkswagen corner
Porsche quietly accumulated control of VW shares and options until the free float was smaller than the shares short-sellers owed. When disclosed, shorts had to buy stock that barely existed; VW briefly became the world’s most valuable company.
Anatomy: the buyer wasn’t willing but obligated — covering is mandatory. Forced flows are the strongest force in markets; our own surviving signal (dealer hedging) is this same anatomy in miniature, every day.
Simons & Medallion
The best-returning fund in history reportedly made its fortune with no hero trades at all: thousands of small, fleeting statistical edges, sized tightly, compounding at extraordinary rates for three decades.
Anatomy: IR = IC × √breadth — a tiny edge taken thousands of times beats a large edge taken once, and it never bets the firm. The greatest “big bet” on this wall is the standing refusal to make one. This museum’s pipeline is built in this shape, not Soros’s.
Long-Term Capital Management
Nobel laureates, genuine edges in convergence trades — levered ~25x and more, on the assumption that historical correlations bound the future. One sovereign default later, the positions were right “eventually” and the fund was gone immediately.
Anatomy, inverted: asymmetry flipped — bounded gains (spreads converge a little), unbounded loss (leverage meets a regime break). Every exhibit above had capped downside; LTCM capped its upside and levered the rest. Markets can stay irrational longer than you can stay solvent — this plaque is why the museum audits regimes.
The Antiquities Wing
the oldest instruments in the collection
Two numbers run every exhibit in this museum. They were both discovered as limits of an honest iteration — which is also this museum’s entire method.
Archimedes bounds π
Archimedes never computed π. He trapped it: a 96-sided polygon inside the circle, another outside, and the honest statement that the truth lives between them — 3 10/71 < π < 3 1/7. Each doubling of the polygon’s sides tightened the bounds. He published the interval, not a point.
Where it hangs in this museum: π sits inside the normal density, so every t-statistic and standard error on every plaque here carries it. But the masterpiece is the method: our survivorship-biased Sharpes are declared as upper bounds; our bootstrap publishes the 5th-to-95th interval, never the point; and the research loop is polygon-doubling — precision is bought by iteration count, not by cleverness per iteration. Grinold’s law says the same thing about breadth: IR = IC × √breadth — edge grows like the square root of the number of honest, independent attempts.
Jacob Bernoulli finds e
Bernoulli asked what happens to money compounded not yearly but continuously: (1 + 1/n)n does not run to infinity — it converges to 2.71828… The same man proved the Law of Large Numbers: frequencies converge to probabilities, but only with enough independent trials.
Where it hangs in this museum: everywhere money grows. Log returns are continuous compounding; the museum’s benchmark line — \$100k in an index fund — is e doing its quiet work over twenty years. The dark twin is volatility drag: compound growth ≈ average return minus half the variance, which is how exhibits with positive average returns still turned \$100k into \$93k. And his Law of Large Numbers is the license for our 10,000-draw bootstrap — with a plaque-sized warning he would have endorsed: it holds for independent draws, and market crises are precisely when independence fails.