LTCM: Why a Nobel-Backed Hedge Fund Handed Back Its Safety Cushion

The room

The psychology: overconfidence

Psychologists split overconfidence into three related errors. Overestimation is thinking you are better than you are. Overplacement is thinking you are better than others. Overprecision is being too sure that your estimates are right, which shows up when people give confidence ranges that are far too narrow. Don Moore and Paul Healy found that the first two come and go with the difficulty of the task, but overprecision is remarkably persistent.

Overprecision is the form that matters most for a leveraged fund, because leverage is a bet on the width of your error bars. If you believe the worst plausible month will cost a few percent of capital, borrowing 28 times your equity looks prudent. If the true range of outcomes is wider than your models say, the same borrowing can wipe you out. The President’s Working Group concluded that the 1998 shock confounded the risk models of LTCM and other firms, and that global portfolios like LTCM’s were less well diversified than assumed. The diversification the models counted on rested on those trades moving fairly independently. In August 1998 they moved together.

None of this shows what any individual partner believed, and the record shows internal debate. What the record does show is a system in which confidence was precise at every level. The fund described its risk as similar to the stock market and uncorrelated with it. Lenders, the GAO found, relied on the principals’ reputations instead of full credit analysis and did not know the size of the positions until days before the rescue. Returning capital while keeping the positions only made sense if the range of bad outcomes was narrow. Alan Greenspan told Congress that sophisticated models can “get too far ahead of human judgment.”

Why smart people fall for it

Three forces turned a defensible balance-sheet decision into a $3.6 billion rescue.

1. 1

   Narrow error bars from short history

   Models fitted to a few calm years treat the worst month in that window as the worst month possible, and ranges come out too tight.

2. 2

   Leverage turns small errors into failure

   At 28 to 1, a fall of under 4 percent in asset values erases the equity, so a modest model error becomes a solvency event.

3. 3

   Reputation stands in for scrutiny

   When lenders and investors trust the people, they ask fewer questions about positions, so no outside check widens the error bars.

Was LTCM’s leverage really the mistake?

The simple story is that LTCM borrowed too much. The fuller record is more interesting. The GAO compared the fund’s balance-sheet leverage of about 28 to 1 with that of large securities firms such as Lehman Brothers and Merrill Lynch and found it broadly similar. Plenty of financial firms run that kind of balance sheet every day. Leverage alone did not make LTCM different.

What made it different was the combination. The fund’s positions were huge relative to the markets it traded, with some futures holdings above 10 percent of open interest, so it could not sell quickly without moving prices against itself. Many of its trades were variations on one bet, that spreads between safer and riskier or less liquid assets would narrow from historically high levels, and in a flight to safety those spreads widen together. Other firms had crowded into the same kinds of trades, so when LTCM needed to sell, the natural buyers were often trying to sell too.

Returning capital in 1997 is where these threads met. The decision did not change a single trade. It changed how much loss the fund could absorb before those trades became a crisis. Alan Greenspan put the general principle bluntly in his testimony: “abnormal returns are sustained only through abnormal exposure to risk.” He added that LTCM had reportedly increased leverage to counter shrinking opportunities, a strategy he judged was destined to fail.

One point is often lost. The consortium held the portfolio, sold it down in an orderly way and got its money back by the end of 1999. That suggests the deeper weakness was less the bets themselves than the fund’s ability to survive long enough to see them through. Being right eventually is worth little if your cushion is too thin to wait.

What does overconfidence research say about LTCM’s models?

Moore and Healy’s experiment is about trivia, not bonds, but its central finding transfers. When people say they are 90 percent sure, they are right much less often than that, and this gap does not disappear as tasks get easier. Their participants were students with a few dollars at stake. Professional risk models are not human guesses, but they are built by humans who choose the data window, the distributions and the assumed correlations, and each of those choices can make a range too narrow.

The President’s Working Group made a version of the same point in institutional language. It warned that risk models based on recent data may understate both the size of shocks and how strongly markets move together during them. That is overprecision written into software: a calm sample produces a tight range, and the tight range justifies more borrowing. The Working Group urged firms to look beyond value-at-risk to stress tests that ask what happens when correlations jump.

Barber and Odean’s study adds the behavioral consequence. Their investors were ordinary households, not hedge funds, but the pattern was clear: the group that research suggests is more prone to overconfidence traded more and kept less of its returns. Overconfidence does not just change what people believe. It changes how much they act on those beliefs, and the cost shows up in results.

Two cautions apply. First, these studies cannot diagnose anyone at LTCM, and the record shows that at least one partner, Myron Scholes, warned internally that firms making similar trades would face failures, according to the Wall Street Journal. Second, confidence was not confined to the fund. Lenders, investors and regulators all accepted a narrow picture of the risks. Overprecision was a property of the whole system, which is why the shock spread so far.

How would a modern risk team decide?

A modern team would start by treating the capital decision and the position decision as one choice. Returning money is never only a question about fees; it is a question about how much loss the remaining equity can absorb. The first number on the table would be how large a fall in asset values would wipe out the fund at the new leverage, set against the worst moves seen in comparable markets over decades, not years.

It would then run stress tests that break the model’s assumptions on purpose. What happens if every spread widens together, as they did in August 1998? What happens if the fund must sell while its competitors sell the same assets? A reverse stress test, which asks what scenario would make the fund fail and how plausible it is, forces the team to describe the disaster in concrete terms instead of a probability.

Third, it would measure liquidity as carefully as price risk. Positions that exceed a meaningful share of a market cannot be exited in days. A modern team would size trades to the time it would take to unwind them in stressed conditions and hold more capital against positions that would take months.

Finally, it would look for outside challenge rather than outside trust. Lenders who see full positions, a risk committee that can veto leverage increases, and a habit of recording forecast ranges and checking how often reality falls outside them all widen the error bars before the market does it for you.

The manager's playbook

1. Link capital decisions to position riskWhenever you return or raise capital, recompute how large a loss would wipe out the remaining equity at the new leverage.
2. Widen your ranges deliberatelyAsk what range you are 90 percent sure of, then check how often past outcomes fell outside similar ranges, and widen accordingly.
3. Stress the correlations, not just pricesTest a scenario in which assets that usually move independently all move against you at once.
4. Size positions to their exit timeEstimate how long it would take to unwind each position in a stressed market and hold more capital against slow exits.
5. Invite outside challenge before trustGive lenders, boards or risk committees enough information to question your assumptions rather than rely on your reputation.

Questions people ask