| Event | Peak | Trough | Drop | Recovery |
|---|---|---|---|---|
| 1929 Depression | Sep 1929: Dow 381 | Jul 1932: Dow 41 | -89% | 25 years |
| 1998 LTCM | $4.7B capital | $0.4B, taken over | -92% | Wound down |
| 2000 Dot-com | Mar 2000: NDX 5048 | Oct 2002: NDX 1114 | -78% | 15 years |
| 2008 GFC | Oct 2007: S&P 1565 | Mar 2009: S&P 676 | -57% | 5.5 years |
1929 was not just a price bubble; it was a collapse of categories. The 1920s mass-marketed speculation: broker margin loans rose from about $3.5B in 1926 to roughly $8.5B by October 1929, letting retail investors capture full upside with 10% of their own money. Households mortgaged their homes to buy Radio Corporation of America (up ~20× from 1925–29) and called it "investing." Graham's three-part definition cut the two apart: ① thorough analysis; ② principal safety (i.e., you can survive the worst case); ③ satisfactory return. Missing any one is speculation — a technical definition, not a moral one.
The Dow peaked at 381.17 on Sep 3, 1929. Black Tuesday (Oct 29, 1929) erased about $14B in market value (nearly 5× that year's federal budget). The index bottomed at 41.22 on Jul 8, 1932 — a 89% drop. The Dow did not regain its 1929 high until November 1954, a full 25 years later. Graham's own partnership lost about 70% between 1929 and 1932 — the disaster that forced him to write Security Analysis (1934) and to make "margin of safety" the spine of the entire system.
The definition itself has edges. ① "Thorough analysis" in fast-moving sectors (early internet, AI, biotech) inherently carries irreducible uncertainty — demanding certainty makes you miss structural opportunities. ② "Principal safety" is relative: in 1933 FDR's repeal of the gold clause amounted to a partial write-down even for Treasury holders. ③ Treating this line as an excuse to "do nothing" is as lazy as chasing rallies. Late in life (1976 interview) Graham himself acknowledged that the marginal payoff of deep single-stock analysis was falling, and pivoted to recommending indexing.
LTCM was founded in 1994 by John Meriwether (former head of Salomon's arbitrage desk), with partners Robert Merton and Myron Scholes — winners of the 1997 Nobel Prize in Economics. The strategy was bond-spread "convergence trades": long illiquid off-the-run Treasuries, short on-the-run, betting tiny spreads would mean-revert. Because spreads were only tens of basis points, they ran direct leverage near 25:1; including OTC derivatives, notional exposure exceeded $1.25 trillion. Models assumed low cross-asset correlation and Gaussian returns — until Russia's August 1998 ruble default triggered a global flight to safety: every spread widened instead of narrowing, every correlation went to 1, and "6-sigma" events occurred repeatedly within four months.
From August to September 1998, LTCM lost about $4.6B, with capital falling from $4.7B to $0.4B. On Sep 23 the New York Fed convened 14 banks to inject $3.625B and take it over, avoiding detonation of the derivatives market. Buffett, Goldman, and AIG had jointly bid $250M cash for the entire book (near a zero clearing price) — rejected by the partners. Merton and Scholes had tens of millions of personal wealth in the fund, almost all wiped out — a Nobel Prize buys no principal safety.
This isn't an argument against all leverage. ① Long-duration, low-cost "benign leverage" (Berkshire's insurance float) has been controllable for decades. ② Not every quantitative strategy is doomed — Renaissance Medallion has run for decades on lower leverage and broader signals. The real distinction: does your leverage carry a forced-liquidation trigger (margin call, mark-to-market)? Leverage that won't be liquidated in the worst historical scenario may be usable; otherwise it's a time bomb — not "if" but "when".
At the March 2000 Nasdaq peak, the weighted P/E (excluding loss-makers) was about 175. The justification was uniform: "we're capturing TAM" — as long as the internet's total addressable market is large enough, today's losses are just "investment in the future." Pets.com IPO'd in Feb 2000 raising $82M and went bankrupt nine months later; Webvan burned through about $1.2B before liquidating in July 2001. But the more insidious story isn't the zeros — it's Cisco: real revenue actually doubled from 2000 to 2010, yet the stock fell from $80 in March 2000 to $8 in 2002 (-90%) and has not regained its 2000 peak more than twenty years later. The company wasn't wrong; the entry price was.
Nasdaq peaked at 5048.62 on Mar 10, 2000 and bottomed at 1114.11 on Oct 9, 2002 — a 77.9% drop that vaporized about $5 trillion in market value. Full recovery took 15 years; the index did not retake 5000 until April 2015. Amazon fell from $113 to $5.51 (-95%) but its fundamentals kept compounding and it became the cycle's biggest winner — yet still took ~9 years to reclaim its 2000 high. The lesson isn't "don't buy tech." It's that the premium you pay for a story, the time you wait for it to deliver, and the drawdown you'll see in between — must all be survivable.
But the bears of 1999 — Shiller, Marks — were "early by several years." Greenspan flagged "irrational exuberance" in December 1996, and shorts missed a doubling of the index over the next four years. ① Overvalued ≠ imminent crash — it can keep rising for years. ② Some "concept" names do eventually deliver (Amazon, Google). ③ In the early phase of a new platform, "growth at any price" can be briefly correct. The real questions: how much did you pay for the story, and how negative is the margin of safety if it doesn't arrive? Damodaran's rule: a valuation story must contain falsifiable assumptions, or it is only a narrative.
2008 wasn't a "subprime corner" collapse — it was the synchronized failure of the entire financial system: AAA-rated CDOs, the repo market, money market funds. Three errors interlocked: ① Hidden leverage: regulatory arbitrage via Citi's and UBS's SIVs/conduits pushed true leverage above 50:1. ② Models assumed regional housing diversification, but from 2007 U.S. home prices fell nationwide together (Case-Shiller dropped ~27% from Jul 2006 to Feb 2012) — the assumption simply didn't hold. ③ Counterparty risk in derivatives was a tangled web — when AIG fell, the entire CDS market froze instantly. The "once in a century" events from VaR models recurred again and again over 18 months — because the models had only ever seen calm data.
The S&P 500 peaked at 1565 on Oct 9, 2007 and bottomed at 676 on Mar 9, 2009 — a 57% drop. Lehman went bankrupt on Sep 15, 2008 ($691B in assets, the largest in U.S. history). AIG received an $85B Fed bridge loan on Sep 16; total rescue funding eventually reached $182.2B. "AAA"-rated subprime MBS suffered actual losses of 60–80%. Bear Stearns was acquired by JPMorgan at $2/share in March 2008 (down from a peak of $172). On Oct 16, 2008 Buffett published "Buy American" in the NYT; in September–October he committed $5B to Goldman and $3B to GE (10% preferred + warrants), yielding about $3B combined profit by 2013 — a generational gift, but only for the few who arrived with cash and a valuation anchor.
The 2008 "bottom fishers" paid prices too. Buffett himself admitted adding to ConocoPhillips when oil was at $140 in 2008 was a "huge mistake" that cost billions. ① "Cheap" in a systemic crash can stay cheap for a long time. ② Trying to time the bottom usually means either missing a deeper bargain or being stranded mid-slope. ③ "Cash is king" only works if you already had cash — raising it inside the crisis is the most expensive form of liquidation. Real preparation happens during calm weather.