When a hedge fund collapses, the story people remember is fraud — a manager who lied, stole, or gambled with client capital. It's the more comfortable explanation, because it implies the failure was a character problem rather than a structural one. A review of 62 major hedge fund blow-ups over 20 years found fraud was the primary cause in only 13 of them, or 21%. The other 79% lost the money legitimately: the strategy simply stopped working, and the business bled out slowly enough that nobody rang an alarm in time.

FX Concepts is a clean example. Once the largest currency hedge fund in the world at roughly $14 billion, its systematic models were built on currency trends and interest-rate differentials that had worked reliably for three decades. Then coordinated quantitative easing after 2009 flattened exactly the signals the models depended on. Assets fell from $12 billion to $661 million by 2013, and the firm filed for bankruptcy. Nothing was stolen. A three-decade edge simply died.

Underneath most of these failures sits the same technical problem: overfitting. A strategy tuned closely enough to historical data will produce an excellent backtest and a live-trading collapse, because it has learned the noise in that specific dataset rather than a durable, repeatable pattern. A widely cited 2014 study showed that the probability of this kind of curve-fitting rises sharply once more than roughly 50 parameter combinations are tested against the same data — a threshold far lower than many strategy development processes actually respect.

It's worth acknowledging the counterpoint: some of the most successful quant funds in history, most famously Renaissance Technologies' Medallion fund, have run concentrated, secretive strategies for decades without visible decay, which seems to argue against the diversify-across-many-strategies approach entirely. But survivorship bias does a lot of work in that comparison — for every Medallion, there are dozens of equally confident, equally sophisticated single-strategy funds that quietly wound down and never became case studies, precisely because nothing dramatic happened; they just stopped being worth writing about. The exceptions prove the model is possible, not that it's the reliable path.

The honest defense against this isn't finding one brilliant strategy and trusting it forever — durable edges decay as more capital crowds into them, and market regimes shift under models trained on the regime that came before. The defense is structural: running enough genuinely uncorrelated strategies that no single one's decay can sink the portfolio, validating each against data it wasn't built on, and rebalancing exposure continuously as conditions change rather than waiting for a drawdown to force the question.

This is a less satisfying story than “the manager was a crook.” It's also the more useful one, because it's the failure mode that's actually waiting for every systematic fund, including disciplined ones — and the only real defense against it is treating strategy decay as the expected cost of doing business, not a surprise to be explained away after the fact.