In the months before its 1998 collapse, Long-Term Capital Management ran a Sharpe ratio of 4.35 — an extraordinary figure, several times what most professional managers would consider excellent. By the standard industry yardstick for risk-adjusted return, LTCM wasn't just good. It looked like one of the best-run funds in the world, staffed by two Nobel laureates in economics and a former vice chairman of the Federal Reserve.

Months later, the fund had lost almost all of its capital and required a Federal Reserve-organized rescue to prevent a broader market seizure. The Sharpe ratio hadn't been lying, exactly — it had been measuring the wrong thing. The metric treats all volatility, upside and downside, as equally undesirable, and it has no mechanism at all for capturing the risk of a rare, correlated shock that hasn't happened yet in the sample period being measured.

LTCM's strategy relied on historical correlations between asset pairs that behaved independently in ordinary markets. When Russia defaulted on its sovereign debt in August 1998, a flight to quality moved nearly every one of those supposedly independent positions in the same direction at once. Historical correlation matrices, built entirely on calm-market data, simply didn't contain the possibility of that kind of simultaneous, stress-driven correlation — because in a genuine stress event, previously uncorrelated assets tend to start moving together, all in response to the same liquidity shock.

In fairness to the metric, William Sharpe never claimed it was a complete risk model — he described it as a simple, single-number heuristic, useful precisely because of how little it demanded of the person using it. The failure at LTCM wasn't that the Sharpe ratio existed; it was that a genuinely brilliant team let a beautiful single number substitute for the harder, less tidy work of stress-testing against scenarios their own historical data had never contained. The tool did exactly what it was built to do. The mistake was asking it to do more than that.

This is the specific failure mode a high Sharpe ratio is structurally blind to: a strategy that earns small, steady, real-looking gains in ordinary conditions while quietly accumulating exposure to a tail event the backtest never saw. Selling insurance against events assumed to be nearly impossible is the textbook version of this trade — premiums flow in steadily, volatility looks low, and the Sharpe ratio looks fantastic, right up until the assumed-impossible event happens.

None of this makes the Sharpe ratio useless — as a quick, single-number gut check across strategies, it still has a place. The mistake is treating it as sufficient on its own. A strategy's maximum historical drawdown, its behavior under simulated stress scenarios that go beyond its own trading history, and its exposure to correlated tail risk all tell you something the Sharpe ratio is built to hide. LTCM's 4.35 remains the permanent reminder of what a beautiful risk-adjusted return number can conceal, right up until the one week it matters most.