Where the model breaks, where it earns its keep

Two pages on Black-Scholes against the real world. Where it failed (LTCM, 1998), where it earns its keep (vanilla options markets quote in implied vol), and where that leaves the question of when to use it.

Long-Term Capital Management (LTCM) was a hedge fund founded in 1994 by John Meriwether, with Nobel laureates Myron Scholes and Robert Merton on its board. The fund pursued convergence trades — long undervalued securities and short overvalued "similar" ones — with leverage of roughly 25–30:1 on-balance-sheet (≈ $30 of borrowed assets controlled per $1 of investor equity) and over $1.25 trillion in notional derivative exposure (the size their swap and futures bets were calibrated to, not the cash they posted) built on just $5 billion of equity. ^·

On August 17, 1998, Russia devalued the ruble and defaulted on its domestic debt, and spreads across nearly every LTCM position diverged instead of converged. The fund lost 44 percent in a single month, and by mid-September its equity was nearly wiped out. ^· An LTCM default would have sent shockwaves through the global banking system, so on September 23, 1998, fourteen banks invested $3.65 billion under Federal Reserve supervision to wind the fund down in an orderly fashion.

The failure traces directly to two Black-Scholes assumptions named in §3.1. First, the continuous-paths assumption broke: the Russian default was a jump, a single discrete event whose impact dwarfed any "normal" daily move. LTCM's models, calibrated to the calm of 1995–1997, treated returns as continuous Gaussian and assigned essentially zero probability to what actually happened. Second, the constant-volatility assumption failed: when σ spiked 5–10× simultaneously across uncorrelated markets, the risk budget built on prior-year volatility was wrong by an order of magnitude — and Jorion (2000) shows LTCM compounded this by using the same covariance matrix to measure and optimize risk, causing the firm to take its largest bets on whatever its miscalibrated model called safest. ^·

In liquid vanilla options markets, traders do not quote option prices in dollars. They quote in implied volatility — the σ that, plugged into Black-Scholes, recovers the market price. A trader observes a price, runs BS in reverse, and the resulting IV becomes the quote. BS functions as a universal translator that maps dollar prices into a single number comparable across strikes and maturities. It is not a truth claim about how markets work.

Why does this work? First, everyone uses it. Every trader and analyst has BS. Counterparties agree on the formula, observe market prices, and exchange IV quotes without negotiating over model choice. A more sophisticated model like Heston or jump-diffusion introduces parameter disputes that slow everything down.

Second, speed. BS prices and Greeks are closed-form, computable in microseconds. A market-maker quoting and re-hedging thousands of strikes at machine speed cannot afford the Fourier inversions or Monte Carlo runs that more realistic models require.

Third, the surface absorbs the wrongness. The skew and term structure named in §3.1 are themselves the system's continuous, market-priced correction layer. BS plus the IV surface gives traders a model they know is wrong, alongside a real-time correction that tells them exactly how it is wrong. That combination is more transparent than any heavier model whose wrongness is hidden inside its parameters. ^·

The defense, then, is that in vanilla liquid markets, a transparently wrong model with a market-updated correction layer beats an opaquely "right" one.

Black-Scholes is not a pricing model. It is an imaginary equation built on imaginary tokens that everyone pretends has meaning, and so it does. Nobody on a real trading desk has believed BS genuinely represents reality since 1987 at the latest, when the post-crash volatility skew became a flashing warning sign above every BS calculator. The reasons why were shown in §3.1: constant σ and continuous paths, both known to be false for decades, and both still baked into the formula. The real model that prices SPX options is the volatility surface — the empirical correction the market layers on top of BS to undo what BS gets wrong. The defense in §4.2 reads as a rescue but actually concedes the point: it defends the convention of quoting in implied volatility, not the formula itself. Calling BS a pricing model at this point is generous. The industry uses BS coordinates because every desk has them, not because the formula is right — BS is a poor system grandfathered in, which cannot be easily removed because the entire financial system is built around it.

The natural pushback is that LTCM proves the danger of mistaking notation for reality, and I agree. §4.1 showed LTCM treating BS-calibrated risk numbers as truth claims at 25:1 leverage, sizing its $5 billion of equity against trillion-dollar notional exposure under the assumption that the model's tails were the market's tails. They were not. The model assigned essentially zero probability to a Russian default, and the market handed LTCM the bill five weeks later. The model was not merely wrong; it was wrong in a direction the IV surface had not yet had time to learn. Therefore, BS is appropriate exactly when you treat it as notation and refuse to treat it as reality. Use it to write the quote. Do not use it to size the position. The industry post-1998 calls this discipline "model risk" and builds whole desks to enforce it.